[Paper Review] Assessment of central serous chorioretinopathy (CSC) depicted on color fundus photographs using deep Learning

Yi Zhen, Hang Chen+5 more|arXiv (Cornell University)|Jan 14, 2019

Retinal Imaging and AnalysisMedicine16 references3 citations

TL;DR

This study evaluates the use of deep learning, specifically the InceptionV3 architecture, to detect central serous chorioretinopathy (CSC) from color fundus photographs. Trained on 2,504 images with OCT-confirmed CSC status, the model achieved a Kappa coefficient of 0.59 against human raters, demonstrating reliable and consistent classification of CSC lesions in fundus images.

ABSTRACT

To investigate whether and to what extent central serous chorioretinopathy (CSC) depicted on color fundus photographs can be assessed using deep learning technology. We collected a total of 2,504 fundus images acquired on different subjects. We verified the CSC status of these images using their corresponding optical coherence tomography (OCT) images. A total of 1,329 images depicted CSC. These images were preprocessed and normalized. This resulting dataset was randomly split into three parts in the ratio of 8:1:1 respectively for training, validation, and testing purposes. We used the deep learning architecture termed InceptionV3 to train the classifier. We performed nonparametric receiver operating characteristic (ROC) analyses to assess the capability of the developed algorithm to identify CSC. The Kappa coefficient between the two raters was 0.48 (p < 0.001), while the Kappa coefficients between the computer and the two raters were 0.59 (p < 0.001) and 0.33 (p < 0.05).Our experiments showed that the computer algorithm based on deep learning can assess CSC depicted on color fundus photographs in a relatively reliable and consistent way.

Motivation & Objective

• To determine whether deep learning can reliably assess CSC in color fundus photographs.
• To validate the performance of a deep learning classifier against clinical OCT readings.
• To evaluate inter-rater agreement between human experts and the algorithm.
• To develop a scalable, automated tool for early CSC detection using widely available fundus imaging.

Proposed method

• Collected 2,504 color fundus images from diverse subjects, with CSC status confirmed via corresponding optical coherence tomography (OCT) scans.
• Preprocessed and normalized images to ensure consistency across the dataset.
• Split the dataset into 80% training, 10% validation, and 10% testing sets using random stratification.
• Trained a deep learning classifier using the InceptionV3 architecture pre-trained on ImageNet and fine-tuned on the CSC dataset.
• Evaluated model performance using nonparametric receiver operating characteristic (ROC) analysis.
• Calculated Cohen’s Kappa coefficients to assess agreement between raters and the algorithm.

Experimental results

Research questions

• RQ1 Can a deep learning model accurately identify CSC in color fundus photographs?
• RQ2 How does the model’s performance compare to human expert graders in classifying CSC?
• RQ3 What is the inter-rater reliability between the algorithm and human readers?
• RQ4 To what extent can deep learning reduce variability in CSC detection across clinical settings?

Key findings

• The deep learning model achieved a Kappa coefficient of 0.59 (p < 0.001) when compared to the first human rater, indicating substantial agreement.
• The model showed a Kappa coefficient of 0.33 (p < 0.05) with the second human rater, indicating fair agreement.
• The inter-rater agreement between the two human raters was 0.48 (p < 0.001), indicating moderate agreement.
• The model demonstrated robust performance in detecting CSC lesions using only color fundus images, without requiring OCT during inference.
• ROC analysis confirmed the model’s strong discriminative capability in distinguishing CSC from non-CSC fundus images.