Deep neural network for early image diagnosis of Stevens-Johnson syndrome/toxic epidermal necrolysis.

Abstract

BACKGROUND\nStevens-Johnson syndrome (SJS)/toxic epidermal necrolysis (TEN) is a life-threatening cutaneous adverse drug reaction (cADR). However, distinguishing SJS/TEN from non-severe cADRS is difficult, especially in the early stages of the disease.\n\n\nOBJECTIVE\nTo overcome this limitation, we developed a computer-aided diagnosis system for the early diagnosis of SJS/TEN, powered by a deep convolutional neural network (DCNN).\n\n\nMETHODS\nWe trained a DCNN using a dataset of 26,661 individual lesion images obtained from 123 patients diagnosed with SJS/TEN or non-severe cADRs. The DCNN s classification accuracy was compared to that of 10 board-certified dermatologists and 24 trainee dermatologists.\n\n\nRESULTS\nThe DCNN achieved 84.6% (95% confidence interval [CI], 80.6－88.6) sensitivity, whereas the sensitivities of the board-certified dermatologists and the trainee dermatologists were 31.3 % (95% CI, 20.9－41.8; P<0.0001) and 27.8% (95% CI, 22.6-32.5; P<0.0001), respectively. The negative predictive value was 94.6% (95% CI, 93.2－96.0) for the DCNN, 68.1 % (95% CI, 66.1－70.0; P<0.0001) for the board-certified dermatologists, and 67.4% (95% CI, 66.1－68.7; P<0.0001) for the trainee dermatologists. The area under the receiver operating characteristic curve of the DCNN for SJS/TEN diagnosis was 0.873, which was significantly higher than that of all the board-certified dermatologists and trainee dermatologists.results CONCLUSIONS: We developed a DCNN to classify SJS/TEN and non-severe cADRs based on individual lesion images of erythema. The DCNN performed significantly better than dermatologists in classifying SJS/TEN from skin images.