The top 4 AUC ROC Probabilities

So the model diagnose an image based(against) on the top 4 performing AUC pathologies, in this case:

  • Cardiomegaly
  • Mass
  • Pneumothorax
  • Edema
    so for our first example on our assignment, the pathology was not in the top 4 probabilities, as a result it results of the above top 4 were all low probabilities.
    My question then is: why can’t the model then show the actual pathology with the highest highest probability? put it differently: why can’t each image load not show the top 4 probabilities of pathologies…and NOT based on the ‘fixed’ top 4 above? I am not sure I make sense.

Hi @Luzuko_Ndzuta,

That’s how the algorithm was built for the assignment. I encourage you to come up with the algorithm based on your suggestions and share your results.