Image-Based Differentiation of Bacterial and Fungal Keratitis Using Deep Convolutional Neural Networks

Abstract

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

The Case for Fungal Keratitis to Be Accepted as a Neglected Tropical Disease

“…There is currently no point of care test for fungal keratitis and this remains a major obstacle in improving health outcomes for the condition. Artificial intelligence (AI) based on deep learning techniques has demonstrated promising performance in detecting microbial keratitis and differentiating between fungal, bacterial and viral causes when using slit-lamp or smartphone images, outperforming even specialist clinicians [ 38 ]. The utilisation of AI could overcome the need for highly trained operators and specialized equipment, which limits the use of many of the other diagnostic alternatives in low-resource settings.…”

Achieving diagnostic excellence for infectious keratitis: A future roadmap

“…In addition, DL-based models have demonstrated good accuracy in differentiating the underlying cause of IK, particularly between bacterial and fungal keratitis (Hung et al, 2021 ; Redd et al, 2022b ), which is a common diagnostic dilemma in clinical practice. Redd et al ( 2022b ) recently evaluated five convolutional neural networks using images of culture-proven IK obtained from handheld cameras and reported a higher diagnostic accuracy of the CNN-based DL models when compared to cornea experts (area under the ROC = 0.84 vs. 0.76).…”

“…In addition, DL-based models have demonstrated good accuracy in differentiating the underlying cause of IK, particularly between bacterial and fungal keratitis (Hung et al, 2021 ; Redd et al, 2022b ), which is a common diagnostic dilemma in clinical practice. Redd et al ( 2022b ) recently evaluated five convolutional neural networks using images of culture-proven IK obtained from handheld cameras and reported a higher diagnostic accuracy of the CNN-based DL models when compared to cornea experts (area under the ROC = 0.84 vs. 0.76). Another study has also shown the ability of a CNN-based DL model in accurately distinguishing between corneal scars and IK (Tiwari et al, 2022 ), which serves as a useful and inexpensive system to aid the triage, assessment, initiation and cessation of antimicrobial therapy for IK in regions with limited access to eye care.…”

Deep Learning Approach in Image Diagnosis of Pseudomonas Keratitis

“…The rapid identification of Pseudomonas keratitis from suspected BK is critical to prevent corneal ulcer patients from corneal melting and blindness. Diagnosing BK from MK via the DL approach of external eye photos has recently shown promise [ 8 , 9 , 10 , 17 ]. Still, there is no report exploring the potential of DL models for further differentiating BK into Pseudomonas and non- Pseudomonas keratitis to date.…”