Improving the Generalizability of Infantile Cataracts Detection via Deep Learning-Based Lens Partition Strategy and Multicenter Datasets

Jiewei Jiang, Shutao Lei, Mingmin Zhu, Ruiyang Li, Jiayun Yue, Jingjing Chen, Zhongwen Li, Jiamin Gong, Duoru Lin, Xiaohang Wu, Zhuoling Lin, Haotian LinFrontiers in Medicine 2021 · DOI

Abstract

Infantile cataract is a leading cause of childhood blindness. While AI diagnostic systems show promise, their generalizability is often limited by the heterogeneity of multicenter slit-lamp images. This study developed two lens partition strategies (LPSs) based on Faster R-CNN and Hough transform to improve detection generalizability. Using 1,643 multicenter images, the Faster R-CNN LPS achieved an average intersection over union (IoU) of 0.9419 and 0.9107 for normal and abnormal lenses, respectively. Compared to the Hough transform, Faster R-CNN improved grading accuracy by 5.31% and reduced processing time from 34.46s to 0.25s per image. This system demonstrates significant potential for real-world clinical application across diverse ophthalmic centers.

Introduction

Infantile cataracts represent a critical global health challenge. Although deep learning models have achieved high performance in controlled, single-center environments, they frequently struggle when deployed across multiple clinics. This lack of generalizability is primarily due to "noise" in slit-lamp images—such as eyelashes, eyelids, and varying lighting conditions—which differs significantly between clinical sites. To overcome this, we propose a standardized Lens Partition Strategy (LPS) to accurately isolate the lens from surrounding ocular noise before diagnostic classification.

Methods

The researchers developed a two-stage framework consisting of an automated lens partition stage followed by a screening and grading stage.

Study Design and Datasets

The study utilized 1,643 slit-lamp images: 886 from the CC-Cruiser dataset (single center) and 757 from a multicenter dataset across four independent clinics. The experimental workflow involved evaluating different LPS methods and determining the minimal multicenter training data required for stable performance.

Overview of the study design, illustrating the evaluation of lens partition strategies and the multicenter data integration workflow.

Technical Architecture

Two LPS methods were compared: the traditional geometric Hough Transform and a deep learning-based Faster R-CNN. The Faster R-CNN model was trained to detect the lens as an object, providing greater robustness against non-circular lens shapes and lighting artifacts. Following partition, a 50-layer ResNet (ResNet-50) was used to perform screening (normal vs. cataract) and three-degree grading (area, density, and location).

Detailed architecture of the Lens Partition Strategy (Faster R-CNN vs. Hough Transform) and the subsequent ResNet-based screening and grading network.

Results

The Faster R-CNN-based LPS significantly outperformed the Hough Transform in both accuracy and efficiency.

  • Accuracy: Faster R-CNN achieved a mean IoU of 0.9419 (normal) and 0.9107 (abnormal), whereas the Hough Transform achieved 0.8970 and 0.8138.
  • Efficiency: Faster R-CNN processed images in 0.25 seconds, a 137x speedup over the 34.46 seconds required by the Hough Transform.
  • Diagnostic Performance: The integrated pipeline achieved an AUC of 0.997 for screening and high AUCs (0.965–0.979) for all grading tasks.
  • Data Requirements: Performance stabilized after adding just 2-3 subsets of multicenter data, indicating that capturing equipment variance is more important than total sample volume.

Discussion

This study demonstrates that the primary bottleneck for AI generalizability in infantile cataract detection is the preprocessing stage. By implementing a robust, deep learning-based lens partition strategy, we can effectively filter out clinical noise and standardize inputs for classification models. While the system is highly effective for whole-lens isolation, future iterations will focus on pixel-level lesion segmentation and the integration of multimodal imaging to capture posterior lens pathologies. This approach paves the way for the deployment of reliable AI diagnostic tools in diverse real-world clinical settings.

Summary updated

Abstract

Infantile cataract is the main cause of infant blindness worldwide. Although previous studies developed artificial intelligence (AI) diagnostic systems for detecting infantile cataracts in a single center, its generalizability is not ideal because of the complicated noises and heterogeneity of multicenter slit-lamp images, which impedes the application of these AI systems in real-world clinics. In this study, we developed two lens partition strategies (LPSs) based on deep learning Faster R-CNN and Hough transform for improving the generalizability of infantile cataracts detection. A total of 1,643 multicenter slit-lamp images collected from five ophthalmic clinics were used to evaluate the performance of LPSs. The generalizability of Faster R-CNN for screening and grading was explored by sequentially adding multicenter images to the training dataset. For the normal and abnormal lenses partition, the Faster R-CNN achieved the average intersection over union of 0.9419 and 0.9107, respectively, and their average precisions are both > 95%. Compared with the Hough transform, the accuracy, specificity, and sensitivity of Faster R-CNN for opacity area grading were improved by 5.31, 8.09, and 3.29%, respectively. Similar improvements were presented on the other grading of opacity density and location. The minimal training sample size required by Faster R-CNN is determined on multicenter slit-lamp images. Furthermore, the Faster R-CNN achieved real-time lens partition with only 0.25 s for a single image, whereas the Hough transform needs 34.46 s. Finally, using Grad-Cam and t-SNE techniques, the most relevant lesion regions were highlighted in heatmaps, and the high-level features were discriminated. This study provides an effective LPS for improving the generalizability of infantile cataracts detection. This system has the potential to be applied to multicenter slit-lamp images.
BibTeX
@inproceedings{jiang2021improving,
  author = {Jiewei Jiang and Shutao Lei and Mingmin Zhu and Ruiyang Li and Jiayun Yue and Jingjing Chen and Zhongwen Li and Jiamin Gong and Duoru Lin and Xiaohang Wu and Zhuoling Lin and Haotian Lin},
  title = {{Improving the Generalizability of Infantile Cataracts Detection via Deep Learning-Based Lens Partition Strategy and Multicenter Datasets}},
  booktitle = {Frontiers in Medicine},
  year = {2021},
  doi = {10.3389/fmed.2021.664023},
}
Figure