Constructing Benchmark Databases and Protocols for Medical Image Analysis: Diabetic Retinopathy

Algorithm 1

Image-wise evaluation based on image scores.

(1)  for  each test image do

(2)  TN←0, TP←0, FN←0, FP←0

(3)  curr_score ← image score

(4)  for each test image do

(5)   if curr_score ≥ image score then

(6)    if ground truth assignment = “normal” then

(7)     TN=TN+1

(8)    else

(9)     FN=FN+1

(10)    end if

(11)   else

(12)    if ground truth assignment = “abnormal”  then

(13)     TP=TP+1

(14)    else

(15)     FP=FP+1

(16)    end if

(17)   end if

(18)  end for

(19)  SN=TPTP+FN (Sensitivity)

(20)  SP=TNTN+FP (Specificity)

(21)   Add new ROC point (𝑥;𝑦) = (1 − SP, SN)

(22)  end for

(23) Return the final ROC curve (all points)