.. _planar_imaging: ============== Planar Imaging ============== Overview -------- .. automodule:: pylinac.planar_imaging :no-members: Feature table ------------- +-------------------+------------------+-------------+----------------+---------------------------+ | Feature/Phantom | Can be inverted? | SSD setting | Auto-centering | Auto-rotation | +===================+==================+=============+================+===========================+ | Doselab MC2 (MV) | No | Manual | Yes | Semi (+/-5 from 0) | +-------------------+------------------+-------------+----------------+---------------------------+ | Doselab MC2 (kV) | No | Manual | Yes | Semi (+/-5 from 0) | +-------------------+------------------+-------------+----------------+---------------------------+ | Las Vegas | No | Manual | Yes | No (0) | +-------------------+------------------+-------------+----------------+---------------------------+ | Elekta Las Vegas | No | Manual | Yes | No (0) | +-------------------+------------------+-------------+----------------+---------------------------+ | Leeds TOR | Yes | Manual | Yes | Yes | +-------------------+------------------+-------------+----------------+---------------------------+ | PTW EPID QC | No | Manual | Yes | No (0) | +-------------------+------------------+-------------+----------------+---------------------------+ | SNC MV | No | Manual | Yes | No (45) | +-------------------+------------------+-------------+----------------+---------------------------+ | SNC MV (12510) | No | Manual | Yes | No (45) | +-------------------+------------------+-------------+----------------+---------------------------+ | SNC kV | No | Manual | Yes | No (135) | +-------------------+------------------+-------------+----------------+---------------------------+ | SI QC-3 (MV) | No | Manual | Yes | Semi (+/-5 from 45/135) | +-------------------+------------------+-------------+----------------+---------------------------+ | SI QC kV | No | Manual | Yes | Semi (+/-5 from 45/135) | +-------------------+------------------+-------------+----------------+---------------------------+ | IBA Primus A | No | Manual | Yes (+/-2cm) | Semi (+/-5 from 0,90,270) | +-------------------+------------------+-------------+----------------+---------------------------+ | ACR Digital Mammo | No | Manual | Yes | No (0) | +-------------------+------------------+-------------+----------------+---------------------------+ .. _typical_planar_usage: Typical module usage -------------------- The following snippets can be used with any of the phantoms in this module; they all have the same or very similar methods. We will use the LeedsTOR for the example, but plug in any phantom from this module. Running the Demo ^^^^^^^^^^^^^^^^ To run the demo of any phantom, create a script or start an interpreter session and input: .. code-block:: python from pylinac import LeedsTOR # or LasVegas, DoselabMC2kV, etc LeedsTOR.run_demo() A figure showing the phantom, low contrast plot, and RMTF will be generated: .. plot:: :include-source: false from pylinac import LeedsTOR LeedsTOR.run_demo() Typical Use ^^^^^^^^^^^ Import the class: .. code-block:: python from pylinac import ( LeedsTOR, ) # or whatever phantom you like from the planar imaging module The minimum needed to get going is to: * **Load image** -- Load the planar image as you would any other class: by passing the path directly to the constructor: .. code-block:: python leeds = LeedsTOR("my/leeds.dcm") Alternatively, a URL can be passed: .. code-block:: python leeds = LeedsTOR.from_url("http://myserver.com/leeds") You may also use the demo image: .. code-block:: python leeds = LeedsTOR.from_demo_image() * **Analyze the images** -- Analyze the image using the :meth:`~pylinac.planar_imaging.LeedsTOR.analyze` method. The low and high contrast thresholds can be specified: .. code-block:: python leeds.analyze(low_contrast_threshold=0.01, high_contrast_threshold=0.5) Additionally, you may specify the SSD of the phantom. By default, SAD and 5cm up from SID are searched: .. code-block:: python leeds.analyze(..., ssd=1400) * **View the results** -- The results of analysis can be viewed with the :meth:`~pylinac.planar_imaging.LeedsTOR.plot_analyzed_image` method. .. code-block:: python leeds.plot_analyzed_image() .. plot:: :include-source: false from pylinac import LeedsTOR lt = LeedsTOR.from_demo_image() lt.analyze(low_contrast_threshold=0.01, high_contrast_threshold=0.5) lt.plot_analyzed_image() Note that each subimage can be turned on or off. .. code-block:: python # don't show the low contrast plot leeds.plot_analyzed_image(low_contrast=False) ROI labels can be enabled to correlate each circle with the corresponding value in the results: .. code-block:: python leeds.plot_analyzed_image(show_roi_labels=True) The figure can also be saved: .. code-block:: python leeds.save_analyzed_image("myprofile.png") A PDF report can also be generated: .. code-block:: python leeds.publish_pdf("leeds_october16.pdf") Leeds TOR Phantom ----------------- The Leeds phantom is used to measure image quality metrics for the kV imager of a linac. It contains both high and low contrast ROIs. .. note:: There are two phantom analysis routines. The :class:`~pylinac.planar_imaging.LeedsTOR` class is for newer phantoms that have a red ring on the outside. Older Leeds phantoms may have a blue label containing the serial number and model on the back. Use the :class:`~pylinac.planar_imaging.LeedsTORBlue` class for these phantoms. The difference is small ROI location shifts. Image Acquisition ^^^^^^^^^^^^^^^^^ You can acquire the images any way you like. Just ensure that the phantom is not touching a field edge. It is also recommended by the manufacturer to rotate the phantom to a non-cardinal angle so that pixel aliasing does not occur for the high-contrast line pairs. Algorithm ^^^^^^^^^ Leeds phantom analysis is straightforward: find the phantom in the image, then sample ROIs at the appropriate locations. The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any size kV imager. * The phantom can be at any distance. * The phantom can be at any angle. * The phantom can be flipped either way. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom must not be touching or close to any image edges. * The blades should be fully or mostly open to correctly invert the image. This may not result in a complete failure, but you may have to force-invert the analysis if this case isn't true (i.e. ``myleeds.analyze(invert=True)``). * The phantom should be centered near the CAX (<1-2cm). **Pre-Analysis** * **Determine phantom location** -- The Leeds phantom is found by performing a Canny edge detection algorithm to the image. The thin structures found are sifted by finding appropriately-sized ROIs. This may include the outer phantom edge and the metal ring just inside. The average central position of the circular ROIs is set as the phantom center. * **Determine phantom angle** -- To find the rotational angle of the phantom, a similar process is employed, but square-like features are searched for in the edge detection image. Because there are two square areas, the ROI with the highest attenuation (lead) is chosen. The angle between the phantom center and the lead square center is set as the angle. * **Determine rotation direction** -- The phantom might be placed upside down. To keep analysis consistent, a circular profile is sampled at the radius of the low contrast ROIs starting at the lead square. Peaks are searched for on each semicircle. The side with the most peaks is the side with the higher contrast ROIs. Analysis is always done counter-clockwise. If the ROIs happen to be clockwise, the image is flipped left-right and angle/center inverted. **Analysis** * **Calculate low contrast** -- Because the phantom center and angle are known, the angles to the ROIs can also be known. The median pixel value of each ROI (see below image, "LC0", ..., "LC17") is compared to the median pixel value of all 4 reference ROIs ("LCR0", ..., "LCR3"). See also :ref:`low_contrast_topic`. By default, the :ref:`michelson` algorithm is used. For example, the first contrast value would be calculated as: .. math:: \frac{LC0_{median} - LCR_{median}}{LC0_{median} + LCR_{median}} * **Calculate high contrast** -- Again, because the phantom position and angle are known, offsets are applied to sample the high contrast line pair regions. For each sample ("HC0", ..., "HC11"), the relative MTF is calculated using Peak-Valley methodology. See :ref:`peak-valley-mtf`. For example, the first high-contrast value would be calculated as: .. math:: \frac{HC0_{max} - HC0_{min}}{HC0_{max} + HC0_{min}} * **Percent Integral Uniformity** -- See :ref:`planar-piu`. **Post-Analysis** * **Determine passing low and high contrast ROIs** -- For each low and high contrast region, the determined value is compared to the threshold. The plot colors correspond to the pass/fail status. See also :ref:`Interpreting Results ` for specific results items. .. plot:: :include-source: false :caption: Labeled ROIs of the Leeds phantom analysis. W/L was adjusted for clarify of the labels. from pylinac import LeedsTOR lt = LeedsTOR.from_demo_image() lt.analyze() lt.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(10, 8), ) Troubleshooting ^^^^^^^^^^^^^^^ If you're having trouble getting the Leeds phantom analysis to work, first check out the :ref:`general_troubleshooting` section. If the issue is not listed there, then it may be one of the issues below. The most common reason for failing is having the phantom near an image edge. The resulting error is usually that the phantom angle cannot be determined. For example, this image would throw an error: .. image:: images/bad_leeds.jpg The below image also fails. Technically, the phantom is in the image, but the top blade skews the pixel values such that the phantom edge cannot be properly found at the top. This fails to identify the true phantom edge, causing the angle to also not be found: .. image:: images/bad_leeds2.jpg Another problem is that the image may have a non-uniform background. This can cause pylinac's automatic inversion correction to incorrectly invert the image. For example, this image falsely inverts: .. image:: images/leeds_uneven.jpg When analyzed, the angle is 180 degrees opposite the lead square, causing the ROIs to be flipped 180 degrees. To correct this problem, pass ``invert=True`` to :meth:`~pylinac.planar_imaging.LeedsTOR.analyze`. This will force pylinac to invert the image the opposite way and correctly identify the lead square. Another common problem is an offset analysis, as shown below: .. image:: images/leeds_offset_inverted.png This is caused by a wrong inversion. .. note:: If the image flash is dark, then the image inversion is very likely wrong. Again, pass ``invert=True`` to the ``analyze`` method. This is the same image but with invert=True: .. image:: images/leeds_offset_corrected.png PTW EPID QC Phantom ------------------- The PTW EPID QC phantom is an MV imaging quality assurance phantom and has high and low contrast regions, just as the Leeds phantom, but with different geometric configurations. .. _epid-qc_image_acquisition: Image Acquisition ^^^^^^^^^^^^^^^^^ The EPID QC phantom appears to have a specific setup as recommended by the manufacturer. The phantom should have the high-contrast line pairs at the top of the image and low contrast at the bottom. The rotation is not automatically determined, so you should take care when setting up the phantom to be well-positioned. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. * The images can be acquired with the phantom at any SSD. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom must be at 0 degrees. * The phantom must not be touching any image edges. * The phantom should have the high-contrast linen pair regions toward the gantry stand/top. * The phantom should be centered near the CAX (<1-2cm). **Pre-Analysis** * **Determine phantom location** -- A Canny edge search is performed on the image. Connected edges that are semi-round and angled are thought to possibly be the phantom. Of the ROIs, the one with the longest axis is said to be the phantom edge. The center of the bounding box of the ROI is set as the phantom center. * **Determine phantom radius** -- The major axis length of the ROI determined above serves as the phantom radius. **Analysis** * **Calculate low contrast** -- Because the phantom center and angle are known, the angles to the ROIs can also be known. The median pixel value of each ROI (see below image, "LC0", ..., "LC8") is compared to the median pixel value of the reference ROI ("LCR"). See also :ref:`low_contrast_topic`. By default, the :ref:`michelson` algorithm is used. For example, the first contrast value would be calculated as: .. math:: \frac{LC0_{median} - LCR_{median}}{LC0_{median} + LCR_{median}} * **Calculate high contrast** -- Again, because the phantom position and angle are known, offsets are applied to sample the high contrast line pair regions. For each sample ("HC0", ..., "HC6"), the relative MTF is calculated using Peak-Valley methodology. See :ref:`peak-valley-mtf`. For example, the first high-contrast value would be calculated as: .. math:: \frac{HC0_{max} - HC0_{min}}{HC0_{max} + HC0_{min}} * **Percent Integral Uniformity** -- See :ref:`planar-piu`. **Post-Analysis** * **Determine passing low and high contrast ROIs** -- For each low and high contrast region, the determined value is compared to the threshold. The plot colors correspond to the pass/fail status. See also :ref:`Interpreting Results ` for specific results items. .. plot:: :include-source: false :caption: Labeled ROIs of the PTW EPID QC phantom analysis. W/L was adjusted for clarify of the labels. from pylinac import PTWEPIDQC ptw = PTWEPIDQC.from_demo_image() ptw.analyze() ptw.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(10, 10), ) Standard Imaging QC-3 Phantom ----------------------------- The Standard Imaging phantom is an MV imaging quality assurance phantom and has high and low contrast regions, just as the Leeds phantom, but with different geometric configurations. .. _qc3_image_acquisition: Image Acquisition ^^^^^^^^^^^^^^^^^ The Standard Imaging phantom has a specific setup as recommended by the manufacturer. The phantom should be angled 45 degrees, with the "1" pointed toward the gantry stand and centered along the CAX. For best results when using pylinac, open the jaws to fully cover the EPID, or at least give 1-2cm flash around the phantom edges. .. warning:: If using the acrylic jig that comes with the phantom, place a spacer of at least a few mm between the jig and the phantom. E.g. a slice of foam on each angled edge. This is because the edge detection of the phantom may fail at certain energies with the phantom abutted to the acrylic jig. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. * The images can be acquired with the phantom at any SSD. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom must be at 45 degrees. * The phantom must not be touching any image edges. * The phantom should have the "1" pointing toward the gantry stand. * The phantom should be centered near the CAX (<1-2cm). **Pre-Analysis** * **Determine phantom location** -- A Canny edge search is performed on the image. Connected edges that are semi-round and angled are thought to possibly be the phantom. Of the ROIs, the one with the longest axis is said to be the phantom edge. The center of the bounding box of the ROI is set as the phantom center. * **Determine phantom radius and angle** -- The major axis length of the ROI determined above serves as the phantom radius. The orientation of the edge ROI serves as the phantom angle. **Analysis** * **Calculate low contrast** -- Because the phantom center and angle are known, the angles to the ROIs can also be known. The median pixel value of each ROI (see below image, "LC0", ..., "LC4") is compared to the median pixel value of the reference ROI ("LCR"). See also :ref:`low_contrast_topic`. By default, the :ref:`michelson` algorithm is used. For example, the first contrast value would be calculated as: .. math:: \frac{LC0_{median} - LCR_{median}}{LC0_{median} + LCR_{median}} * **Calculate high contrast** -- Again, because the phantom position and angle are known, offsets are applied to sample the high contrast line pair regions. For each sample ("HC0", ..., "HC4"), the relative MTF is calculated using Peak-Valley methodology. See :ref:`peak-valley-mtf`. For example, the first high-contrast value would be calculated as: .. math:: \frac{HC0_{max} - HC0_{min}}{HC0_{max} + HC0_{min}} * **Percent Integral Uniformity** -- See :ref:`planar-piu`. **Post-Analysis** * **Determine passing low and high contrast ROIs** -- For each low and high contrast region, the determined value is compared to the threshold. The plot colors correspond to the pass/fail status. See also :ref:`Interpreting Results ` for specific results items. .. plot:: :include-source: false :caption: Labeled ROIs of the Standard Imaging QC-3 phantom analysis. W/L was adjusted for clarify of the labels. from pylinac import StandardImagingQC3 qc3 = StandardImagingQC3.from_demo_image() qc3.analyze() qc3.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(10, 8), ) Troubleshooting ^^^^^^^^^^^^^^^ If you're having issues with the StandardImaging class, make sure you have correctly positioned the phantom as per the manufacturer's instructions (also see :ref:`qc3_image_acquisition`). One issue that may arise is incorrect inversion. If the jaws are closed tightly around the phantom, the automatic inversion correction may falsely invert the image, just as for the Leeds phantom. If you have an image that looks inverted or just plain weird, add ``invert=True`` to :meth:`~pylinac.planar_imaging.StandardImagingQC3.analyze`. If this doesn't help, reshoot the phantom with the jaws open. Standard Imaging QC-kV Phantom ------------------------------ The Standard Imaging QC-kV phantom is an kV imaging quality assurance phantom and has high and low contrast regions, just as the Leeds phantom, but with different geometric configurations. Image Acquisition ^^^^^^^^^^^^^^^^^ The Standard Imaging phantom has a specific setup as recommended by the manufacturer. The phantom should be angled 45 degrees, with the "1" pointed toward the gantry stand and centered along the CAX. For best results when using pylinac, open the blades to fully cover the kV panel, or at least give 1-2cm flash around the phantom edges. .. warning:: If using the acrylic jig that comes with the phantom, place a spacer of at least a few mm between the jig and the phantom. E.g. a slice of foam on each angled edge. This is because the edge detection of the phantom may fail at certain energies with the phantom abutted to the acrylic jig. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any kV panel. * The images can be acquired with the phantom at any SSD. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom must be at 45 degrees. * The phantom must not be touching any image edges. * The phantom should have the "1" pointing toward the gantry stand. * The phantom should be centered near the CAX (<1-2cm). **Pre-Analysis** * **Determine phantom location** -- A Canny edge search is performed on the image. Connected edges that are semi-round and angled are thought to possibly be the phantom. Of the ROIs, the one with the longest axis is said to be the phantom edge. The center of the bounding box of the ROI is set as the phantom center. * **Determine phantom radius and angle** -- The major axis length of the ROI determined above serves as the phantom radius. The orientation of the edge ROI serves as the phantom angle. **Analysis** * **Calculate low contrast** -- Because the phantom center and angle are known, the angles to the ROIs can also be known. The median pixel value of each ROI (see below image, "LC0", ..., "LC4") is compared to the median pixel value of the reference ROI ("LCR"). See also :ref:`low_contrast_topic`. By default, the :ref:`michelson` algorithm is used. For example, the first contrast value would be calculated as: .. math:: \frac{LC0_{median} - LCR_{median}}{LC0_{median} + LCR_{median}} * **Calculate high contrast** -- Again, because the phantom position and angle are known, offsets are applied to sample the high contrast line pair regions. For each sample ("HC0", ..., "HC4"), the relative MTF is calculated using Peak-Valley methodology. See :ref:`peak-valley-mtf`. For example, the first high-contrast value would be calculated as: .. math:: \frac{HC0_{max} - HC0_{min}}{HC0_{max} + HC0_{min}} **Post-Analysis** * **Determine passing low and high contrast ROIs** -- For each low and high contrast region, the determined value is compared to the threshold. The plot colors correspond to the pass/fail status. See also :ref:`Interpreting Results ` for specific results items. .. plot:: :include-source: false :caption: Labeled ROIs of the QC-kV phantom analysis. from pylinac import StandardImagingQCkV qckv = StandardImagingQCkV.from_demo_image() qckv.analyze() qckv.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(10, 8), ) Troubleshooting ^^^^^^^^^^^^^^^ If you're having issues with the StandardImaging class, make sure you have correctly positioned the phantom as per the manufacturer's instructions (also see :ref:`qc3_image_acquisition`). One issue that may arise is incorrect inversion. If the jaws are closed tightly around the phantom, the automatic inversion correction may falsely invert the image, just as for the Leeds phantom. If you have an image that looks inverted or just plain weird, add ``invert=True`` to :meth:`~pylinac.planar_imaging.StandardImagingQCkV.analyze`. If this doesn't help, reshoot the phantom with the jaws open. Las Vegas Phantom ----------------- The Las Vegas phantom is for MV image quality testing and includes low contrast regions of varying contrast and size. There is also a :class:`~pylinac.planar_imaging.ElektaLasVegas` class that is very similar. This section covers both styles. Image Acquisition ^^^^^^^^^^^^^^^^^ The Las Vegas phantom has a recommended position as stated on the phantom. Pylinac will however account for shifts and inversions. Best practices for the Las Vegas phantom: * Keep the phantom from a couch edge or any rails. * The field edge should be >=5mm from the phantom edge, preferably 10+mm. * The orientation should have the largest "holes" towards the right side although this can be accounted for as an ``analyze`` parameter. * The angle should be as close to 0 as possible, given above, although this can be accounted for as an ``analyze`` parameter. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom must not be touching any image edges. * The phantom should be at a cardinal angle (0, 90, 180, or 270 degrees) relative to the EPID. * The phantom should be centered near the CAX (<1-2cm). **Pre-Analysis** * **Determine phantom location** -- A Canny edge search is performed on the image. Connected edges that are semi-round and angled are thought to possibly be the phantom. Of the ROIs, the one with the longest axis is said to be the phantom edge. The center of the bounding box of the ROI is set as the phantom center. * **Determine phantom radius and angle** -- The major axis length of the ROI determined above serves as the phantom radius. The orientation of the edge ROI serves as the phantom angle. **Analysis** * **Calculate low contrast** -- Because the phantom center and angle are known, the angles to the ROIs can also be known. The median pixel value of each ROI (see below image, "LC0", ..., "LC18") is compared to the median pixel value of all 4 reference ROIs ("LCR0", ..., "LCR3"). See also :ref:`low_contrast_topic`. By default, the :ref:`michelson` algorithm is used. For example, the first contrast value would be calculated as: .. math:: \frac{LC0_{median} - LCR_{median}}{LC0_{median} + LCR_{median}} **Post-Analysis** * **Determine passing low and high contrast ROIs** -- For each low and high contrast region, the determined value is compared to the threshold. The plot colors correspond to the pass/fail status. See also :ref:`Interpreting Results ` for specific results items. .. plot:: :include-source: false :caption: Labeled ROIs of the Las Vegas phantom analysis. W/L was adjusted for clarify of the labels. from pylinac import LasVegas lv = LasVegas.from_demo_image() lv.analyze() lv.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(10, 10), ) Doselab MC2 MV & kV ------------------- The Doselab MC2 phantom is for both kV & MV image quality testing and includes low and high contrast regions of varying contrast. There are two high contrast sections, one intended for kV and one for MV. Image Acquisition ^^^^^^^^^^^^^^^^^ The Doselab phantom has a recommended position as stated on the phantom. Pylinac will however account for shifts and inversions. Best practices for the Doselab phantom: * Keep the phantom away from a couch edge or any rails. * Center the phantom along the CAX. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom must not be touching any image edges. * The phantom should be at 45 degrees relative to the EPID. * The phantom should be centered near the CAX (<1-2cm). **Pre-Analysis** * **Determine phantom location** -- A canny edge search is performed on the image. Connected edges that are semi-round and angled are thought to possibly be the phantom. Of the ROIs, the one with the longest axis is said to be the phantom edge. The center of the bounding box of the ROI is set as the phantom center. * **Determine phantom radius and angle** -- The major axis length of the ROI determined above serves as the phantom radius. The orientation of the edge ROI serves as the phantom angle. **Analysis** * **Calculate low contrast** -- Because the phantom center and angle are known, the angles to the ROIs can also be known. The median pixel value of each ROI (see below image, "LC0", ..., "LC6") is compared to the median pixel value of the reference ROI ("LCR"). See also :ref:`low_contrast_topic`. By default, the :ref:`michelson` algorithm is used. For example, the first contrast value would be calculated as: .. math:: \frac{LC0_{median} - LCR_{median}}{LC0_{median} + LCR_{median}} * **Calculate high contrast** -- Again, because the phantom position and angle are known, offsets are applied to sample the high contrast line pair regions. For each sample ("HC0", ..., "HC3"), the relative MTF is calculated using Peak-Valley methodology. See :ref:`peak-valley-mtf`. For example, the first high-contrast value would be calculated as: .. math:: \frac{HC0_{max} - HC0_{min}}{HC0_{max} + HC0_{min}} **Post-Analysis** * **Determine passing low and high contrast ROIs** -- For each low and high contrast region, the determined value is compared to the threshold. The plot colors correspond to the pass/fail status. See also :ref:`Interpreting Results ` for specific results items. .. plot:: :include-source: false :caption: Labeled ROIs of the Doselab MC2 kV phantom analysis. W/L was adjusted for clarify of the labels. from pylinac import DoselabMC2kV dl_kv = DoselabMC2kV.from_demo_image() dl_kv.analyze() dl_kv.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(10, 8), ) .. plot:: :include-source: false :caption: Labeled ROIs of the Doselab MC2 MV phantom analysis. W/L was adjusted for clarify of the labels. from pylinac import DoselabMC2MV dl_mv = DoselabMC2MV.from_demo_image() dl_mv.analyze() dl_mv.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(10, 10), ) SNC MV & kV ----------- The SNC MV and kV phantoms are for kV & MV image quality testing and includes low and high contrast regions of varying contrast. Image Acquisition ^^^^^^^^^^^^^^^^^ The SNC phantoms typically use the angled setup jig. Best practices for the Doselab phantom: * Keep the phantom away from a couch edge or any rails. * Center the phantom along the CAX. * Use the angled setup jig. * For the MV phantom, have the longer side point inferiorly (i.e. **away** from the stand). * For the kV phantom, have the longer side point superiorly (i.e. **toward** the stand). Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom must not be touching any image edges. * The phantom should be at 45 degrees relative to the EPID. * The phantom should be centered near the CAX (<1-2cm). **Pre-Analysis** * **Determine phantom location** -- A canny edge search is performed on the image. Connected edges that are semi-round and angled are thought to possibly be the phantom. Of the ROIs, the one with the longest axis is said to be the phantom edge. The center of the bounding box of the ROI is set as the phantom center. * **Determine phantom radius** -- The major axis length of the ROI determined above serves as the phantom radius. **Analysis** * **Calculate low contrast** -- Because the phantom center and angle are known, the angles to the ROIs can also be known. The median pixel value of each ROI (see below image, "LC0", ..., "LC3") is compared to the median pixel values of both of the reference ROIs ("LCR0", "LCR1"). See also :ref:`low_contrast_topic`. By default, the :ref:`michelson` algorithm is used. For example, the first contrast value would be calculated as: .. math:: \frac{LC0_{median} - LCR_{median}}{LC0_{median} + LCR_{median}} * **Calculate high contrast** -- Again, because the phantom position and angle are known, offsets are applied to sample the high contrast line pair regions. For each sample ("HC0", ..., "HC3"), the relative MTF is calculated using Peak-Valley methodology. See :ref:`peak-valley-mtf`. For example, the first high-contrast value would be calculated as: .. math:: \frac{HC0_{max} - HC0_{min}}{HC0_{max} + HC0_{min}} **Post-Analysis** * **Determine passing low and high contrast ROIs** -- For each low and high contrast region, the determined value is compared to the threshold. The plot colors correspond to the pass/fail status. See also :ref:`Interpreting Results ` for specific results items. .. plot:: :include-source: false :caption: Labeled ROIs of the SNC kV phantom analysis. W/L was adjusted for clarify of the labels. from pylinac import SNCkV snc_kv = SNCkV.from_demo_image() snc_kv.analyze() snc_kv.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(10, 8), ) .. plot:: :include-source: false :caption: Labeled ROIs of the SNC MV phantom analysis. W/L was adjusted for clarify of the labels. from pylinac import SNCMV snc_mv = SNCMV.from_demo_image() snc_mv.analyze() snc_mv.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(10, 10), ) .. _iba_primus_a: IBA Primus A ------------ The IBA Primus A phantom is used for kV image analysis and includes low and high contrast regions of varying contrast. Image Acquisition ^^^^^^^^^^^^^^^^^ Lay the phantom on the couch with the wedge step circle facing the top/gun and high-res square facing the bottom/target. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom must not be touching any image edges. * The phantom should be at 0, 90, or 270 +/-5 degrees relative to the EPID where 0 is facing the gun. * The dynamic wedge steps should be facing the gun side; the high-resolution square should be facing the target side. * The phantom should be centered near the CAX (<2cm). **Pre-Analysis** * **Determine phantom location** -- A Canny edge search is performed on the image. The ROI that approximates the size of the central crosshair of the phantom and is nearly at the center of the image is used as the phantom center location * **Determine phantom radius** -- The size of the above crosshair ROI is used as the basis for the phantom radius. * **Fine-tune phantom angle** -- The phantom angle is assumed to be around 0 (wedge steps facing gun), but fine-tuning is performed so that sensitive ROIs like MTF can be had with high accuracy. This is performed by taking a circular profile about the phantom at the radius of the wedge steps. The two areas of highest gradient will be at the first and last wedge steps. The center between these two points is the angle at which the phantom is "pointing" and will be used as the updated angle. .. warning:: If the gradients cannot be found or if the determined angle is >5 degrees (caused by bad inversion, e.g.) a warning will be printed to the console and a default of 0 will be used. **Analysis** * **Calculate low contrast** -- Because the phantom center and angle are known, the angles to the ROIs can also be known. The median pixel value of each ROI (see below image, "LC0", ..., "LC14") is compared to the median pixel values of the reference ROI ("LCR"). See also :ref:`low_contrast_topic`. By default, the :ref:`michelson` algorithm is used. For example, the first contrast value would be calculated as: .. math:: \frac{LC0_{median} - LCR_{median}}{LC0_{median} + LCR_{median}} * **Calculate high contrast** -- Again, because the phantom position and angle are known, offsets are applied to sample the high contrast line pair regions. For each sample ("HC0", ..., "HC12"), the relative MTF is calculated using Peak-Valley methodology. See :ref:`peak-valley-mtf`. For example, the first high-contrast value would be calculated as: .. math:: \frac{HC0_{max} - HC0_{min}}{HC0_{max} + HC0_{min}} **Post-Analysis** * **Determine passing low and high contrast ROIs** -- For each low and high contrast region, the determined value is compared to the threshold. The plot colors correspond to the pass/fail status. See also :ref:`Interpreting Results ` for specific results items. .. plot:: :include-source: false :caption: Labeled ROIs of the IBA Primus A phantom analysis. from pylinac import IBAPrimusA primus = IBAPrimusA.from_demo_image() primus.analyze(ssd=1395) primus.plot_analyzed_image( show_roi_labels=True, low_contrast=False, high_contrast=False, figsize=(8, 10), ) .. _acr_digital_mammography: ACR Digital Mammography ----------------------- .. warning:: This phantom analysis is still in beta. The parameters may change in future releases. The ACR Digital Mammography phantom is for testing the performance of Full-Field Digital Mammography (FFDM) systems. Image Acquisition ^^^^^^^^^^^^^^^^^ The ACR Digital Mammography phantom has a recommended position as stated on `ACR Phantom Testing: Mammography (Revised 11-22-2024) `__ Best practices for the ACR Digital Mammography phantom: * Place the chest-wall side of phantom completely flush with chest-wall side of image receptor. * Small deviations from being flush can be accounted for using ``angle_adjustment``, ``x_adjustment``, and ``y_adjustment`` in the ``analyze`` parameter. Algorithm ^^^^^^^^^ **Pre-Analysis** * **Determine phantom location** -- A Canny edge search is performed on the image. Connected edges that are semi-round and angled are thought to possibly be the phantom. Of the ROIs, the one closest to the expected area is said to be the phantom edge. The center of the bounding box of the ROI is set as the phantom center. **Analysis** The ACR Digital Mammography phantom is composed of 3 test objects: Masses, Speck groups, and Fibers. * **Calculate masses** -- Because the phantom center and angle are known, the angles to the ROIs can also be known. The median pixel value of each ROI is compared to the median pixel value of the reference ROIs. See also :ref:`low_contrast_topic`. By default, the :ref:`michelson` algorithm is used. For example, the first contrast value would be calculated as: .. math:: \frac{LC0_{median} - LCR_{median}}{LC0_{median} + LCR_{median}} * **Calculate speck groups** -- Because the phantom center and angle are known, the positions to the center of the speck groups can also be known. Furthermore, since the individual specks within a group follow the vertices of a pentagon, their positions can also be know. The visibility of the individual specks is evaluated with a variation of the :ref:`visibility` algorithm .. math:: Visibility(I, R) = Contrast(I, R) * \frac{\sqrt{\pi * radius^2}}{R_{std}} where ``I`` is the max intensity in the ROI, ``R`` is the mean signal of the ROI, :math:`R_{std}` is the standard deviation of the ROI, and ``radius`` is the radius of the speck per manual description. By default, the ``Weber`` algorithm is used for ``Contrast``. * **Calculate fibers** -- Because the phantom center and angle are known, the positions to the fibers can also be known. A fiber is detected using a sequence of image processing filters: #. Vesselness filter -- uses the Frangi filter to detect vessel-like structures. #. Thresholding -- convert to binary. #. Closing -- connects parts of a fiber if there are breaks. #. Label -- detect connected blobs. #. Regionprops -- detect the blob most likely to be the fiber. #. Length -- calculate the length of the fiber as the length of the major axis of the ellipse in regionprops. Standard Imaging FC-2 --------------------- The FC-2 phantom is for testing light/radiation coincidence. Image Acquisition ^^^^^^^^^^^^^^^^^ The FC-2 phantom should be placed on the couch at 100cm SSD. * Keep the phantom away from a couch edge or any rails. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom should be at a cardinal angle (0, 90, 180, or 270 degrees) relative to the EPID. * The phantom should be centered near the CAX (<1cm). * The phantom should be +/- 1cm from 100cm SSD. **Pre-Analysis** * **Determine BB set to use** -- There are two sets of BBs, one for 10x10cm and another for 15x15cm. To get the maximum accuracy, the larger set is used if a 15x15cm field is irradiated. The field size is determined and if it's >14cm then the algorithm will look for the larger set. Otherwise, it will look for the smaller 4. **Analysis** * **Get BB centroid** -- Once the BB set is chosen, image windows look for the BBs in a 1x1cm square. Once it finds them, the centroid of all 4 BBs is calculated. * **Determine field center** -- The field size is measured along the center of the image in the inplane and crossplane direction. A 5mm strip is averaged and used to reduce noise. **Post-Analysis** * **Comparing centroids** -- The irradiated field centroid is compared to the EPID/image center as well as the the BB centroid. The field size is also reported. .. _doselab_rlf: Doselab RLf ----------- .. versionadded:: 3.15 The Doselab RLf is for testing light/radiation coincidence. See also :class:`~pylinac.planar_imaging.DoselabRLf`. Image Acquisition ^^^^^^^^^^^^^^^^^ The RLf phantom should be placed on the couch at 100cm SSD. * Keep the phantom away from a couch edge or any rails. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom should be at a cardinal angle (0, 90, 180, or 270 degrees) relative to the EPID. * The phantom should be centered near the CAX (<2mm). **Analysis** * **Get BB centroid** -- An image window looks for each BB on the inner side of each edge. After finding the BBs, the centroid is calculated. .. note:: The inner 10x10 BBs are always used regardless of the field size. This is because the BB detection is more robust when the BBs are away from a field edge. This also means that 10x10 analysis is slightly less robust that 15x15 analysis all else being equal. * **Determine field center** -- The field size is measured along the center of the image in the inplane and crossplane direction. A 5mm strip is averaged and used to reduce noise. **Post-Analysis** * **Comparing centroids** -- The irradiated field centroid is compared to the EPID/image center as well as the the BB centroid. The field size is also reported. .. _isoalign: IsoAlign -------- .. versionadded:: 3.15 The IsoAlign phantom is for testing light/radiation coincidence. See also :class:`~pylinac.planar_imaging.IsoAlign`. Image Acquisition ^^^^^^^^^^^^^^^^^ The phantom should be placed on the couch at 100cm SSD. * Keep the phantom away from a couch edge or any rails. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom should be at a cardinal angle (0, 90, 180, or 270 degrees) relative to the EPID. * The phantom should be centered near the CAX (<2mm). **Analysis** * **Get BB centroid** -- An image window looks for the central BB as well as 1 BB in each cardinal direction. After finding the BBs, the centroid is calculated. * **Determine field center** -- The field size is measured along the center of the image in the inplane and crossplane direction. A 10mm strip is averaged and used to reduce noise. **Post-Analysis** * **Comparing centroids** -- The irradiated field centroid is compared to the EPID/image center as well as the the BB centroid. The field size is also reported. .. _imt_lrad: IMT L-Rad --------- .. versionadded:: 3.2 The IMT L-Rad phantom is for testing light/radiation coincidence. Unlike the FC-2 phantom, the L-Rad's BBs are all the way at the edge of the phantom. This means for any size below 20x20cm those BBs can't be seen. Even at 20x20, the field edge partially obscures the BBs. For this reason, we only use the central BB for detection. Image Acquisition ^^^^^^^^^^^^^^^^^ The L-Rad phantom should be placed on the couch at 100cm SSD. * Keep the phantom away from a couch edge or any rails. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom should be at a cardinal angle (0, 90, 180, or 270 degrees) relative to the EPID. * The phantom should be centered near the CAX (<3mm). **Analysis** * **Get BB centroid** -- An image window looks for the central BB in a 1.2x1.2cm square. Once it finds it, the centroid is calculated. * **Determine field center** -- The field size is measured along the center of the image in the inplane and crossplane direction. A 5mm strip is averaged and used to reduce noise. **Post-Analysis** * **Comparing centroids** -- The irradiated field centroid is compared to the EPID/image center as well as the the BB centroid. The field size is also reported. .. _snc-fsqa: SNC FSQA -------- .. versionadded:: 3.3 The SNC FSQA phantom is for light/radiation coincidence. It contains markers which guide the physicist on how to position the light field for either a 10x10 or 15x15cm field. There is also an offset BB 4cm at the top right of the image. Because of both :ref:`the philosophy ` of pylinac on light/rad and also because pylinac is a library and not a GUI, there is no interaction to find the edge markers. Instead, we use the one offset BB and then offset that point back 4cm in each direction to get a "virtual center". This center is compared to the field center and EPID center. The expectation is that the physicist set up their field to the markers using the light field at the time of acquisition. Image Acquisition ^^^^^^^^^^^^^^^^^ The FSQA phantom should be placed on the couch at 100cm SSD. * Keep the phantom away from a couch edge or any rails. * Keep the phantom as close to 0 degrees rotation as possible. Algorithm ^^^^^^^^^ The algorithm works like such: **Allowances** * The images can be acquired at any SID. * The images can be acquired with any EPID. **Restrictions** .. warning:: Analysis can fail or give unreliable results if any Restriction is violated. * The phantom should be at 0 degrees relative to the EPID. * The phantom should be roughly centered along the CAX (<3mm). **Analysis** * **Get BB centroid** -- An image window looks for the top-right offset BB in a 1.2x1.2cm square. Once it finds it, a "virtual center" centroid is calculated by shifting the detected BB location by 4cm in each direction. * **Determine field center** -- The field size is measured along the center of the image in the inplane and crossplane direction. A 5mm strip is averaged and used to reduce noise. **Post-Analysis** * **Comparing centroids** -- The irradiated field centroid is compared to the EPID/image center as well as the the BB centroid. The field size is also reported. Analysis Parameters ------------------- .. tab-set:: :sync-group: usage .. tab-item:: pylinac :sync: pylinac See the ``analyze`` method of the class. E.g. :meth:`pylinac.planar_imaging.LasVegas.analyze`. .. tab-item:: RadMachine :sync: radmachine * **Source-to-Phantom distance**: The distance in mm from the phantom to the source. * **Normalized high contrast threshold**: The rMTF minimum value for a region to be considered passing. * **Angle override**: The angle in degrees to override the automatic angle determination. * **Contrast definition**: The method to use to calculate contrast. See :ref:`contrast`. * **Contrast threshold**: The minimum contrast value for a region to be considered passing. Analysis Parameters (Light/Rad) ------------------------------- These are the analysis parameters for Light/Rad phantoms. .. tab-set:: :sync-group: usage .. tab-item:: pylinac :sync: pylinac See :meth:`pylinac.planar_imaging.IMTLRad.analyze` and similar classes for details. .. tab-item:: RadMachine :sync: radmachine * **FWXM height**: The percent height of the maximum to use as the field width. * **BB edge distance threshold**: The threshold in mm to determine if the BB is near the edge of the image. If the BB-to-field-edge is less than this threshold, a different, more robust algorithm is used to determine the BB position but results in higher uncertainty when in a flat region (i.e. away from the field edge). Analysis Parameters (ACR Mammography) ------------------------------------- These are the analysis parameters for ACR Digital Mammography phantom: .. tab-set:: :sync-group: usage .. tab-item:: pylinac :sync: pylinac See :meth:`pylinac.planar_imaging.ACRDigitalMammography.analyze` for details. .. tab-item:: RadMachine :sync: radmachine * **Source-to-Phantom distance**: The distance in mm from the phantom to the source. * **Mass Contrast definition**: The method to use to calculate contrast for mass ROIs. See :ref:`contrast`. * **Mass Visibility threshold**: The minimum contrast value for a mass ROI to be considered passing. * **Speck Group Contrast definition**: The method to use for calculating the contrast of speck groups. * **Speck Group Visibility threshold**: The threshold for whether a speck is "seen". * **Speck group half threshold**: The speck group score is 0.5 if the number of visible specks is between this value and the full threshold. * **Speck group full threshold**: The speck group score is 1.0 if the number of visible specks is larger or equal than this value. * **Max Fiber gap**: The max allowed gap in mm between partial fibers. * **Fiber length half threshold**: The fiber score is 0.5 if its length is between this value and the full threshold (in mm). * **Fiber length full threshold**: The fiber score is 1.0 if its length is larger or equal than this value (in mm). * **Fiber orientation tolerance**: The tolerance in degrees to validate the fiber orientation. * **X adjustment**: A fine-tuning adjustment to the detected x-coordinate of the phantom center in mm. Positive values move the phantom to the right. * **Y adjustment**: A fine-tuning adjustment to the detected y-coordinate of the phantom center in mm. Positive values move the phantom down. * **Angle adjustment**: A fine-tuning adjustment to the detected angle of the phantom in degrees. Positive values rotate the phantom clockwise. * **ROI size factor**: A fine-tuning adjustment to the ROI sizes. This scales the ROIs by this amount. Positive values increase the ROI sizes. * **Scaling factor**: A fine-tuning adjustment to the detected magnification of the phantom. This zooms the ROIs and phantom outline by this amount. Interpreting Results -------------------- .. _interpreting-planar-results: Phantoms ^^^^^^^^ The results from phantoms that are meant to measure image quality, contrast, etc, are: * ``median_contrast``: The median contrast of the low contrast ROIs. See :ref:`low_contrast_topic` and :ref:`cnr` for more. * ``median_cnr``: The median contrast-to-noise ratio of the low contrast ROIs. * ``num_contrast_rois_seen``: The number of low contrast ROIs that had a visibility score above the passed threshold. See :ref:`visibility` for more. * ``phantom_center_x_y``: The center of the phantom in the image in pixels. * ``phantom_area``: The area of the phantom in pixels^2. * ``mtf_lp_mm``: The 80%, 50%, and 30% MTF values in lp/mm. For more values see: :ref:`calculate-specific-mtf`. * ``percent_integral_uniformity``: The percent integral uniformity of the image. See :ref:`planar-piu`. * ``low_contrast_rois``: A dictionary of the individual low contrast ROIs. The dictionary keys are the ROI number, starting at 0. Each ROI has the following information: * ``contrast method``: The method used to calculate the contrast. See :ref:`low_contrast_topic`. * ``contrast``: The contrast value of the ROI. * ``visibility``: The visibility score of the ROI. * ``visibility threshold``: The threshold used to determine visibility. * ``cnr``: The contrast-to-noise ratio of the ROI. * ``passed visibility``: Whether the ROI passed the visibility threshold. * ``signal to noise``: The signal-to-noise ratio of the ROI. .. _interpreting-light-rad-results: Light/Rad ^^^^^^^^^ Light/radiation coincidence phantoms are used to ensure that the light field and radiation field are aligned. The results from these analyses are: * ``field_size_x_mm``: The size of the field in the x-direction/crossplane in mm. * ``field_size_y_mm``: The size of the field in the y-direction/inplane in mm. * ``field_epid_offset_x_mm``: The offset of the field center from the EPID/image center in the x-direction/crossplane in mm. * ``field_epid_offset_y_mm``: The offset of the field center from the EPID/image center in the y-direction/inplane in mm. * ``field_bb_offset_x_mm``: The offset of the field center from the BB center in the x-direction/crossplane in mm. * ``field_bb_offset_y_mm``: The offset of the field center from the BB center in the y-direction/inplane in mm. .. note:: Some phantoms have multiple BBs. When we speak of the BB center when multiple BBs are present, we are referring to the centroid of all BBs. .. _planar-piu: Percent Integral Uniformity ^^^^^^^^^^^^^^^^^^^^^^^^^^^ The uniformity of the phantom can be found by using the :meth:`~pylinac.planar_imaging.LeedsTOR.percent_integral_uniformity` method. This uses the same equation as ACR for CT uniformity. See the "Uniformity" section under :ref:`ACR Analysis ` for more information. The PIU is calculated over all the low-contrast ROIs and the lowest (worst) PIU is returned. For robustness, the 1st and 99th percentiles are used rather than the min/max. The true min/max can be influenced by salt and paper noise. To use the true min and max, set the percentiles to 0 and 100 respectively: .. note:: Calls to ``percent_integral_uniformity`` will not be reflected in the ``results_data`` structure; it will always use (1, 99). .. code-block:: python leeds = LeedsTOR(...) leeds.analyze(...) print(leeds.percent_integral_uniformity(percentiles=(0, 100))) # uses the true min/max .. warning:: This equation was chosen because it is common and understood, but it does come with pitfalls. It is not designed to handle negative values or 0. The calculated result may be misleading if these conditions exist. Customizing Light/Rad behavior ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ The BB window as well as the expected BB positions, and field strip size can be overridden like so: .. code-block:: python from pylinac import StandardImagingFC2 # works for any light/rad phantom class MySIFC2(StandardImagingFC2): bb_sampling_box_size_mm = ( 20 # look at a 20x20mm window for the BB at the expected position ) # change the 10x10 BB expected positions. This is in mm relative to the CAX. bb_positions_10x10 = { "TL": [-30, -30], "BL": [-30, 30], "TR": [30, -30], "BR": [30, 30], } bb_positions_15x15 = ... # same as above field_strip_width_mm = 10 # 10mm strip in x and y to determine field size # use as normal fc2 = MySIFC2(...) .. _creating_a_custom_phantom: Creating a custom planar phantom -------------------------------- In the event you would like to analyze a phantom that pylinac does not analyze out of the box, the pylinac planar imaging module structure allows for generating new phantom analysis types quickly and easily. The benefit of this design is that with a few simple definitions you inherit a strong base of methods (e.g. plotting and PDF reports come for free). Creating a new class involves a few different steps but can be done in a few minutes. The following is a guide for custom phantoms: 1. Subclass the ``ImagePhantomBase`` class: .. code-block:: python from pylinac.planar_imaging import ImagePhantomBase class CustomPhantom(ImagePhantomBase): pass 2. Define the ``common_name``. This is the name shown in plots and PDF reports. .. code-block:: python class CustomPhantom(ImagePhantomBase): common_name = "Custom Phantom v2.0" 3. If the phantom has a high-contrast measurement object, define the ROI locations. .. code-block:: python class CustomPhantom(ImagePhantomBase): ... high_contrast_roi_settings = { "roi 1": { "distance from center": 0.5, "angle": 30, "roi radius": 0.05, "lp/mm": 0.2, }, # add as many ROIs as are needed } .. note:: The exact values of your ROIs will need to be empirically determined. This usually involves an iterative process of adjusting the values until the values are satisfactory based on the ROI sample alignment to the actual ROIs. 4. If the phantom has a low-contrast measurement object, define the sample ROI and background ROI locations. .. code-block:: python class CustomPhantom(ImagePhantomBase): ... low_contrast_roi_settings = { "roi 1": { "distance from center": 0.5, "angle": 30, "roi radius": 0.05, }, # no lp/mm key # add as many ROIs as are needed } low_contrast_background_roi_settings = { "roi 1": {"distance from center": 0.3, "angle": -45, "roi radius": 0.02}, # add as many ROIs as are needed } .. note:: The exact values of your ROIs will need to be empirically determined. This usually involves an iterative process of adjusting the values until the values are satisfactory based on the ROI sample alignment to the actual ROIs. 5. Set the "detection conditions", which is the list of rules that must be true to properly detect the phantom ROI. E.g. the phantom should be near the center of the image. Detection conditions must always have a specific signature as shown below: .. code-block:: python def my_special_detection_condition( region: RegionProperties, instance: object, rtol: float ) -> bool: # region is a scikit regionprop (https://scikit-image.org/docs/dev/api/skimage.measure.html#skimage.measure.regionprops) # instance == self of the phantom # rtol is relative tolerance of agreement. Don't have to use this. do_stuff # e.g. is the region size and position correct? return bool(result) # must always return a boolean class CustomPhantom(ImagePhantomBase): detection_conditions = [ my_special_detection_condition, ] # list of conditions; add as many as you want. 6. Optionally, add a phantom outline object. This helps visualize the algorithm's determination of the size, center, and angle. If no object is defined, then no outline will be shown. This step is optional. .. code-block:: python class CustomPhantom(ImagePhantomBase): ... phantom_outline_object = { "Circle": {"radius ratio": 0.5} } # to create a circular outline # or... phantom_outline_object = { "Rectangle": {"width ratio": 0.5, "height ratio": 0.3} } # to create a rectangular outline At this point you could technically call it done. You would need to always override the angle, center, and size values in the analyze method however. To automate this part you will need to fill in the associated logic. You can use whatever method you like. What I have found most useful is to use an edge detection algorithm and find the outline of the phantom. .. code-block:: python class CustomPhantom(ImagePhantomBase): ... def _phantom_center_calc(self) -> Point: # do stuff in here to determine the center point location. # don't forget to return as a Point item (pylinac.core.geometry.Point). ... def _phantom_radius_calc(self) -> float: # do stuff in here to return a float that represents the phantom radius value. # This value does not have to relate to a physical measure. It simply defines a value that the ROIs scale by. ... def _phantom_angle_calc(self) -> float: # do stuff in here to return a float that represents the angle of the phantom. # Again, this value does not have to correspond to reality; it simply offsets the ROIs. # You may also return a constant if you like for any of these. ... Congratulations! You now have a fully-functioning custom phantom. By using the base class and the predefined attributes and methods, the plotting and PDF report functionality comes for free. Usage tips, tweaks, & troubleshooting ------------------------------------- .. _fine-tuning-planar: Fine-tuning the ROI locations ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ .. versionadded:: 3.24 If after the automatic analysis you find that the ROIs are not quite where you want them, you can adjust the ROI locations by setting any of the following parameters: ``x_adjustment``, ``y_adjustment``, ``angle_adjustment``, ``scaling_factor``, or ``zoom_factor``. These parameters can be set in the ``analyze`` method. .. code-block:: python from pylinac import LeedsTOR leeds = LeedsTOR(...) leeds.analyze( ..., x_adjustment=0.5, y_adjustment=-0.3, angle_adjustment=5, scaling_factor=1.1, roi_size_factor=0.9, ) In contrast to the ``angle_override``, ``size_override``, and ``center_override`` parameters, the adjustments are applied **after** the phantom localization. I.e. use adjustments if you need to fine-tune the automatic analysis; use overrides if the detection is failing. Set the SSD of your phantom ^^^^^^^^^^^^^^^^^^^^^^^^^^^ If your phantom is at a non-standard distance (!= 1000mm), e.g. sitting on the EPID panel, you can specify its distance via the ``ssd`` parameter. .. warning:: The ``ssd`` should be in mm, not cm. Pylinac is moving toward consistent units on everything and it will be mm for distance. .. code-block:: python from pylinac import StandardImagingQC3 qc = StandardImagingQC3(...) qc.analyze(..., ssd=1500) # distance to the phantom in mm. Adjust an ROI on an existing phantom ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ .. note:: If you are trying to uniformly adjust all the ROIs, see :ref:`fine-tuning-planar`. To adjust an ROI, override the relevant attribute or create a subclass. E.g. to move the 2nd ROI of the high-contrast ROI set of the QC-3 phantom: .. code-block:: python from pylinac import StandardImagingQC3 StandardImagingQC3.high_contrast_roi_settings["roi 1"][ "distance from center" ] = 0.05 # overrides that one setting qc3 = StandardImagingQC3(...) # or class TweakedStandardImagingQC3(StandardImagingQC3): high_contrast_roi_settings = { "roi 1": ... } # note that you must replace ALL the values qc3 = TweakedStandardImagingQC3(...) .. _calculate-specific-mtf: Calculate a specific MTF ^^^^^^^^^^^^^^^^^^^^^^^^ To calculate a specific MTF value, i.e. the frequency at a given MTF%: .. code-block:: python dl = DoselabMC2kV(...) dl.analyze(...) print(dl.mtf.relative_resolution(x=50)) # 50% rMTF Get/View the contrast of a low-contrast ROI ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ .. code-block:: python leeds = LeedsTOR(...) leeds.analyze(...) print(leeds.low_contrast_rois[1].contrast) # get the 2nd ROI contrast value Loosen the ROI finding conditions ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If for some reason you have a need to loosen the existing phantom-finding algorithm conditions you can do so fairly easily by overloading the current tooling: .. code-block:: python from pylinac.planar_imaging import is_right_size, is_centered, LeedsTOR def is_right_size_loose(region, instance, rtol=0.3): # rtol default is 0.1 return is_right_size(region, instance, rtol) # set the new condition for whatever LeedsTOR.detection_conditions = [is_right_size_loose, is_centered] # proceed as normal myleeds = LeedsTOR(...) .. _planar_scaling: Scaling measurement ------------------- .. versionadded:: 3.19 Pylinac can produce an area calculation of the phantom. This can be used as a way to test the scaling of the imager per TG-142. The scaling is based on the blue rectangle/circle that is shown in the plots. E.g.: .. code-block:: python leeds = pylinac.LeedsTOR(...) leeds.analyze(...) results = leeds.results_data() print(results.phantom_area) # in mm^2 .. warning:: The produced scaling value is based on the blue rectangle/circle. In many cases it does not equal the exact size of the phantom. It is recommended to be used as a constancy check. Adjusting the scaling ^^^^^^^^^^^^^^^^^^^^^ .. note:: This can also be adjusted uniformly using the ``scaling_factor`` parameter in the ``analyze`` method. The below method is recommended if your adjustments are not uniform in both directions. See :ref:`fine-tuning-planar`. If you are dead-set on having the scaling value be the exact size of the phantom, or you simply have a different interpretation of what the scaling should be you can override the scaling calculation to a degree. The scaling is calculated using the ``phantom_outline_object`` attribute. This attribute is a dictionary and defines the size of the rectangle/circle that is shown in the plots. Changing these values will both change the plot and the area/scaling value. .. code-block:: python import pylinac class NewSNCkV(pylinac.SNCkV): phantom_outline_object = { "Rectangle": {"width ratio": 8.4, "height ratio": 7.2} # change these } class NewLeeds(pylinac.LeedsTOR): phantom_outline_object = {"Circle": {"radius ratio": 1.3}} # change this Wrong phantom angle ------------------- It may sometimes be that the angle of the phantom appears incorrect, or the results appear incorrect. E.g. here is a QC-3 phantom: .. image:: images/bad_qc3.png The ROIs appear correct, the but the contrast and MTF do not monotonically decrease, indicating a problem. In this case, it is because the image acquisition rules were not followed. For the QC-3, the "1" should always point toward the gantry, as per the manual. When oriented this way, the results will be correct. .. _light-rad-philosophy: Light/Radiation philosophy -------------------------- Pylinac (or rather the author) has an opinionated philosophy about light vs radiation that affects the way light/radiation analysis is performed. In our opinion, light/rad **using a phantom** is antiquated as EPIDs are robust enough nowadays to be quite reliable, at least for Varian machines. By using something as simple as graph paper after mechanical measurements, a light field can be set and a simple open field delivered. This open field size and CAX offset can be compared to the nominal values set by the physicist at the time of acquisition. Short of using CCD cameras or specialty equipment like phosphorus, there is no true way to know the light field measurement. All we have is what the physicist set up to. If the physicist sets up to a nominal size like 10x10, then a radiation field measurement can be compared to that rather simply with common field analysis. E.g if the measured field size was 10.1x10.6mm then the error between light and rad is 0.1 and 0.6mm respectively. The CAX offset follows the same logic. You may disagree, but this is here for the purposes of explaining our philosophy and why light/rad does (or does not do) what it does. We provide these light/rad routines because customers ask for them, not because we recommend them. API Documentation ----------------- .. autoclass:: pylinac.planar_imaging.LeedsTOR :inherited-members: .. autoclass:: pylinac.planar_imaging.LeedsTORBlue :inherited-members: .. autoclass:: pylinac.planar_imaging.StandardImagingQC3 :inherited-members: .. autoclass:: pylinac.planar_imaging.StandardImagingQCkV :inherited-members: .. autoclass:: pylinac.planar_imaging.LasVegas :inherited-members: .. autoclass:: pylinac.planar_imaging.ElektaLasVegas :inherited-members: .. autoclass:: pylinac.planar_imaging.DoselabMC2MV :inherited-members: .. autoclass:: pylinac.planar_imaging.DoselabMC2kV :inherited-members: .. autoclass:: pylinac.planar_imaging.SNCMV :inherited-members: .. autoclass:: pylinac.planar_imaging.SNCMV12510 :inherited-members: .. autoclass:: pylinac.planar_imaging.SNCkV :inherited-members: .. autoclass:: pylinac.planar_imaging.PTWEPIDQC :inherited-members: .. autoclass:: pylinac.planar_imaging.IBAPrimusA :inherited-members: .. autoclass:: pylinac.planar_imaging.StandardImagingFC2 :inherited-members: .. autoclass:: pylinac.planar_imaging.IMTLRad :inherited-members: .. autoclass:: pylinac.planar_imaging.DoselabRLf :inherited-members: .. autoclass:: pylinac.planar_imaging.IsoAlign :inherited-members: .. autoclass:: pylinac.planar_imaging.SNCFSQA :inherited-members: .. autoclass:: pylinac.planar_imaging.ACRDigitalMammography :inherited-members: .. autopydantic_model:: pylinac.planar_imaging.ACRDigitalMammographyResult .. autopydantic_model:: pylinac.planar_imaging.PlanarResult