Digital Image Processing and Development of Machine Learning Models for the Discrimination of Corneal Pathology: An Experimental Model

Andres Bustamante-Arias,Abbas Cheddad,Alejandro Rodriguez-Garcia,Julio Cesar Jimenez-Perez

doi:10.3390/photonics8040118

Andres Bustamante-Arias, Abbas Cheddad + Show 2 more

Open Access

https://doi.org/10.3390/photonics8040118

Copy DOI

Abstract

Machine learning (ML) has an impressive capacity to learn and analyze a large volume of data. This study aimed to train different algorithms to discriminate between healthy and pathologic corneal images by evaluating digitally processed spectral-domain optical coherence tomography (SD-OCT) corneal images. A set of 22 SD-OCT images belonging to a random set of corneal pathologies was compared to 71 healthy corneas (control group). A binary classification method was applied where three approaches of ML were explored. Once all images were analyzed, representative areas from every digital image were also extracted, processed and analyzed for a statistical feature comparison between healthy and pathologic corneas. The best performance was obtained from transfer learning—support vector machine (TL-SVM) (AUC = 0.94, SPE 88%, SEN 100%) and transfer learning—random forest (TL- RF) method (AUC = 0.92, SPE 84%, SEN 100%), followed by convolutional neural network (CNN) (AUC = 0.84, SPE 77%, SEN 91%) and random forest (AUC = 0.77, SPE 60%, SEN 95%). The highest diagnostic accuracy in classifying corneal images was achieved with the TL-SVM and the TL-RF models. In image classification, CNN was a strong predictor. This pilot experimental study developed a systematic mechanized system to discern pathologic from healthy corneas using a small sample.

Highlights

In recent years, there has been a significant advancement in computerized corneal digital imaging analysis to develop objective and reproducible machine learning (ML)algorithms for preclinical detection and measurement of various corneal pathologic changes.Standardized quantitative measurement of different corneal structural alterations, such as stromal thinning and edema, inflammatory infiltration, fibrosis and scarring, are crucial for early detection, objective documentation, grading, disease progression, and therapeutic monitoring.The great diversity of unspecific corneal pathologic lesions, and their significant overlap between disorders, represent a major disadvantage for analysis with spectral-domain optical coherence tomography (SD-OCT) [1]
Supervised Machine learning (ML) has been applied to systematic identification and diagnosis of different ocular pathologies, including diabetic retinopathy [5,6], age-related macular degeneration [7,8,9,10], glaucoma [11,12,13], keratoconus [14,15,16,17], corneal edema [18]
A total of 93 SD-OCT corneal images were registered in the study, 71 images formed part of the control group, and 22 pathologic images were included in the experimental group

Summary

Introduction

There has been a significant advancement in computerized corneal digital imaging analysis to develop objective and reproducible machine learning (ML)algorithms for preclinical detection and measurement of various corneal pathologic changes.Standardized quantitative measurement of different corneal structural alterations, such as stromal thinning and edema, inflammatory infiltration, fibrosis and scarring, are crucial for early detection, objective documentation, grading, disease progression, and therapeutic monitoring.The great diversity of unspecific corneal pathologic lesions, and their significant overlap between disorders, represent a major disadvantage for analysis with spectral-domain optical coherence tomography (SD-OCT) [1]. Standardized quantitative measurement of different corneal structural alterations, such as stromal thinning and edema, inflammatory infiltration, fibrosis and scarring, are crucial for early detection, objective documentation, grading, disease progression, and therapeutic monitoring. Unlike other corneal imaging technologies, including corneal topography-tomography and aberrometry that analyze numerical data, the SD-OCT has difficulty providing precise measurement values over varied and unspecific morphologic patterns that could guide clinicians to more objective diagnostic analysis [1,2]. The latter represents a significant challenge for AI developers. Supervised ML has been applied to systematic identification and diagnosis of different ocular pathologies, including diabetic retinopathy [5,6], age-related macular degeneration [7,8,9,10], glaucoma [11,12,13], keratoconus [14,15,16,17], corneal edema [18]

Objectives

Results

Conclusion