Cataract on ultrasound....

Deep Learning-Based Classification of Canine Cataracts from Ocular B-Mode Ultrasound Images

Sanghyeon Park 1, Seokmin Go 2, Seonhyo Kim 3, Jaeho Shim 4 5

Background
Cataracts are a common ocular disorder in dogs, leading to vision impairment or blindness. While B-mode ocular ultrasonography is a widely used, accessible imaging modality in veterinary practice, its diagnostic interpretation is subjective and depends heavily on the examiner's expertise. Traditional computer-aided diagnosis methods required manual feature engineering and lacked scalability. The study aimed to develop and evaluate convolutional neural network (CNN)-based models to automatically classify canine cataracts from ocular ultrasound images into four stages: No cataract, Cortical, Mature, and Hypermature.

Methods
The researchers utilized a dataset of 3,155 B-mode ultrasound images from the AI-HUB platform and external veterinary clinics. The dataset was divided into training (80%), validation (10%), and test (10%) sets while preserving class distribution. Data augmentation techniques and class-weighted loss functions were employed to address class imbalance. Four CNN architectures (AlexNet, EfficientNet-B3, ResNet-50, DenseNet-161) were trained using transfer learning with ImageNet pre-trained weights. Model training involved fine-tuning all layers using the Adam optimizer over 100 epochs, with early stopping applied.

Results
DenseNet-161 achieved the best classification performance, with a test accuracy of 92.03% and an F1 score of 0.8744 on the combined test set. External validation confirmed this performance, yielding 92.15% accuracy and an F1 score of 0.9231. The highest classification accuracy was observed for the "No cataract" class (99.0%), while the "Hypermature" class had the lowest accuracy (78.6%). ROC analysis showed AUC values of 0.99 for DenseNet-161 and ResNet-50, confirming high discriminative power. Grad-CAM visualizations demonstrated that the model consistently focused on the lens region across all categories.

Limitations
The dataset had class imbalances, particularly for Hypermature cataracts, which may have affected model performance. Image acquisition protocols and ultrasound equipment were not standardized. While the model achieved high accuracy, it cannot replace comprehensive clinical evaluations that include additional clinical information. Furthermore, the model’s reliance on single-label image-level annotations may limit its granularity in feature interpretation.

Conclusions
This study presents a deep learning-based method for the automatic classification of canine cataracts using ocular ultrasound images. The DenseNet-161 model showed robust performance and strong generalizability, highlighting its potential as a clinical decision support tool in veterinary ophthalmology. The approach could be extended to other ocular pathologies in future research, thereby enhancing diagnostic capabilities in resource-limited settings.

Disclaimer: The summary generated in this email was created by an AI large language model. Therefore errors may occur. Reading the article is the best way to understand the scholarly work. The figure presented here remains the property of the publisher or author and subject to the applicable copyright agreement. It is reproduced here as an educational work. If you have any questions or concerns about the work presented here, reply to this email.