A deep learning approach in diagnosing fungal keratitis based on corneal photographs

Abstract

Fungal keratitis (FK) is the most devastating and vision-threatening microbial keratitis, but clinical diagnosis a great challenge. This study aimed to develop and verify a deep learning (DL)-based corneal photograph model for diagnosing FK. Corneal photos of laboratory-confirmed microbial keratitis were consecutively collected from a single referral center. A DL framework with DenseNet architecture was used to automatically recognize FK from the photo. The diagnoses of FK via corneal photograph for comparing DL-based models were made in the Expert and NCS-Oph group through a majority decision of three non-corneal specialty ophthalmologist and three corneal specialists, respectively. The average percentage of sensitivity, specificity, positive predictive value, and negative predictive value was approximately 71, 68, 60, and 78. The sensitivity was higher than that of the NCS-Oph (52%, P < .01), whereas the specificity was lower than that of the NCS-Oph (83%, P < .01). The average accuracy of around 70% was comparable with that of the NCS-Oph. Therefore, the sensitive DL-based diagnostic model is a promising tool for improving first-line medical care at rural area in early identification of FK.

Figure 1

The deep learning framework for diagnosing fungal keratitis by corneal photographs. (a) The abstract structure for developing the deep learning-based model for diagnosing fungal keratitis. (b) The DenseNet architecture, a deep learning neural network adopted in this study. Images were fed into the first convolution layer and then the output feature was mapped to input the dense block. Dense blocks contain dense networks that connect each layer to every other layer in a feedforward manner. The output of first two dense blocks were the input of transition layers that reduce the dimensions of the channels to prevent further dense blocks from generating too many feature maps. The last dense block produced feature maps, and these maps were fed in the global average pooling layer, fully connected layer, and Softmax to obtain the final classification results. DL, deep learning; ReLU, rectified linear unit; FC, fully connected layer.

Figure 2

Representative photographs of microbial keratitis caused by fungal and non-fungal pathogens. Fungal keratitis (a–d) and non-fungal keratitis (e–h): (a) Candida keratitis, (b) Fusarium keratitis, (c) Acremonium keratitis, (d) Curvularia keratitis, (e) Pseudomonas keratitis, (f) Herpes keratitis, (g) Acanthamoeba keratitis, (h) Microsporidia keratitis.

Figure 3

The performance of the deep learning model in differentiating fungal keratitis and non-fungal keratitis was illustrated by the receiver operator characteristic curve. ROC receiver operator characteristic curve plot.

Figure 4

Diagnostic performance of the deep learning model and senior ophthalmologists in the identification of cases of fungal keratitis from total 288 photographs of microbial keratitis. (a) Comparing the DL-based model with the NCS-Oph diagnosis. (b) Comparing the DL-based model with the Expert diagnosis. DL deep learning, NCS-Oph non-corneal specialty ophthalmologists, PPV positive predictive value, NPV negative predictive value; each box was constructed by five parameters, including the mean (center of box), lower and upper 90% confidence limits (floor and top of box), and lower and upper 95% confidence limits (lower and upper error bars). P < 0.05 was recognized as statistical difference and determined by two-tailed Fisher exact test.