Receiver Operating Characteristic Curves (ROC Curves):



  1. ROC curves are used to assess the usefulness of a test for detecting something (e.g. cancer).
  2. When a diagnostic test is performed, there will always be a gray area where the positive group and negative group overlap to a certain extent.  This makes medicine difficult.
  3. We must talk about a few outcomes of decision making (a decision threshold for a test is represented by the vertical line above). Depending on where the decision threshold is placed: 
  1. TP - true positive
  2. FP - false positive
  3. TN - true negative
  4. FN - false negative 
  1. There are a number of definitions that can be defined using the above nomenclature: 
  1. True Positive Fraction (TPF): This is the sensitivity of the test, calling something positive when it is:


  1. False Positive Fraction (FPF): calling something positive when it’s not:


  1. Specificity: Calling something normal when it is:


  1. Accuracy: Calling something what it truly is:


  1. A ROC curve (shown above on right above) is generated for a number of different decision thresholds.  The further it bends towards the upper left corner the better the test is (the straight line is a 50/50 guess). A decision threshold should be picked such that it balances the number of false positives with the number of true positives.


kV, MV, Pencil Beam, Fan Beam, Cone beam and Image Quality:

  1. We encounter many different ways of imaging a patient that utilizes all different energies and different beam geometries.  They all influence the final image quality.
  2. kV versus MV imaging: 
  1. kV images offer the best contrast between soft tissue and bony anatomy.
  2. While kV energies may have significant Compton interactions, a large proportion of interactions are still photoelectric and the effective Z of bone is nearly 2x that of tissue (remember the photoelectric effect is proportional to Z3).
  3. MV energies exhibit virtually no photoelectric interactions.  Therefore, we only measure attenuation due to Compton interactions (we can still see an image from a port film due to the large doses involved, compare MV = 1 R and kV = 20 mR).
  4. kV imaging sometimes cannot penetrate a thick person’s anatomy (especially in the lateral direction).  In these cases, MV imaging offers the ability to actually see an image.  Also, MV imaging can be used to see if a prosthesis is hollow. 
  1. Beam shape: 
  1. Scatter degrades our images.  Any Compton scattering events that occur will ideally scatter away from the detector and not be counted.   But, this does not always occur.
  2. Beam geometry changes the amount of scatter degrading your image.  For instance: 
  1. A pencil beam (think 1st generation CT) has no scatter.  Any scattering events send the photon away from the detector.
  2. Fan beams scan faster but have scattered components within the beam that degrade the image.
  3. Cone beams scan the fastest but have scatter in both dimensions that degrades image quality. 
  1. Most modern CT scanners utilize a cone beam of some sort; however, the size of the cone is important.
  2. For instance, on a LINAC taking a CBCT uses a square field with lots of scatter and hence lower image quality than a diagnostic CT that uses a thin rectangular beam. 
  1. Scatter can be rejected using a grid as discussed above, but that only works for kV imaging as the grid becomes too large and heavy for MV imaging.


2019-05-26, 16:53
ABRPhysicsHelp provided an excellent outline to study for part 3. The material was detailed enough to give a good overview required to prepare for the exam. For a deeper understanding of subjects ABRPhysicsHelp would suggest reference material to review. Also the mock oral exam section I found helpful in organizing my thoughts and then talking out the answers all under the pressure of a time limit.