[Paper Review] Comparative Performance Analysis of Transformer-Based Pre-Trained Models for Detecting Keratoconus Disease

Nayeem Ahmed, Md Maruf Rahman+4 more|arXiv (Cornell University)|Aug 16, 2024

Corneal surgery and disorders5 citations

TL;DR

The paper compares eight pre-trained CNNs to detect keratoconus and identifies MobileNetV2 as the most accurate model, with InceptionV3 and DenseNet121 also performing well, while some models struggle with borderline cases.

ABSTRACT

This study compares eight pre-trained CNNs for diagnosing keratoconus, a degenerative eye disease. A carefully selected dataset of keratoconus, normal, and suspicious cases was used. The models tested include DenseNet121, EfficientNetB0, InceptionResNetV2, InceptionV3, MobileNetV2, ResNet50, VGG16, and VGG19. To maximize model training, bad sample removal, resizing, rescaling, and augmentation were used. The models were trained with similar parameters, activation function, classification function, and optimizer to compare performance. To determine class separation effectiveness, each model was evaluated on accuracy, precision, recall, and F1-score. MobileNetV2 was the best accurate model in identifying keratoconus and normal cases with few misclassifications. InceptionV3 and DenseNet121 both performed well in keratoconus detection, but they had trouble with questionable cases. In contrast, EfficientNetB0, ResNet50, and VGG19 had more difficulty distinguishing dubious cases from regular ones, indicating the need for model refining and development. A detailed comparison of state-of-the-art CNN architectures for automated keratoconus identification reveals each model's benefits and weaknesses. This study shows that advanced deep learning models can enhance keratoconus diagnosis and treatment planning. Future research should explore hybrid models and integrate clinical parameters to improve diagnostic accuracy and robustness in real-world clinical applications, paving the way for more effective AI-driven ophthalmology tools.

Motivation & Objective

• Motivate automated keratoconus diagnosis using deep learning and evaluate multiple pre-trained CNN architectures on a curated dataset.
• Assess how well each model separates keratoconus, normal, and suspicious cases using standardized training settings.
• Identify strengths and weaknesses of state-of-the-art CNNs for keratoconus identification to guide future model development.

Proposed method

• Evaluate eight pre-trained CNNs on a keratoconus dataset with three classes: keratoconus, normal, and suspicious.
• Apply consistent training configurations across models, including bad sample removal, resizing, rescaling, and augmentation.
• Measure performance using accuracy, precision, recall, and F1-score to compare class separation capabilities.

Experimental results

Research questions

• RQ1 Which pre-trained CNN architectures provide the highest accuracy in keratoconus detection on the selected dataset?
• RQ2 How do different models perform on ambiguous/suspicious cases compared to clear keratoconus or normal cases?
• RQ3 What are the relative strengths and weaknesses of popular CNN backbones for automated keratoconus diagnosis?

Key findings

• MobileNetV2 achieves the best accuracy in identifying keratoconus and normal cases with few misclassifications.
• InceptionV3 and DenseNet121 perform well for keratoconus detection but show difficulties with questionable/suspicious cases.
• EfficientNetB0, ResNet50, and VGG19 have more trouble distinguishing dubious cases from regular ones, indicating a need for refinement.
• The study highlights varying benefits and weaknesses across state-of-the-art CNN architectures for automated keratoconus identification.