New AI hybrid detects cataract severity with 95% accuracy using smartphone photos
No GPU or special hardware needed—just a regular camera phone and 95% accuracy.
A team led by K. Mithra and Prem Kumar Santhanam has published a paper (arXiv:2607.18349) detailing a low-cost, automated cataract severity classification system that runs on standard consumer-grade color photographs of the eye. The hybrid framework combines deep features extracted from a Convolutional Neural Network (CNN) with five handcrafted Grey-Level Co-occurrence Matrix (GLCM) and intensity descriptors—mean intensity, uniformity, standard deviation, contrast, and energy. A Hough-circle transform localizes the pupil region of interest (ROI) before feature fusion. A multi-class Support Vector Machine (SVM) with an RBF kernel then categorizes each image into one of four severity grades: normal, immature, mature, or hypermature cataract.
The approach was validated on a dataset of 300 ophthalmologist-labeled images (75 per class) collected at an ophthalmology clinic. The fused CNN-GLCM-SVM system achieved 95.0% accuracy, 93.8% sensitivity, and 96.1% specificity, significantly outperforming the texture-only baseline (88.5%) and CNN-only baseline (91.3%). It also surpassed recently published deep learning methods. The key advantage: the system requires no GPU acceleration or specialized ophthalmic cameras, making it suitable for primary-care and telemedicine deployment in resource-limited settings. This work could bring affordable cataract screening to underserved populations, enabling early detection and treatment.
- Fuses CNN deep features with 5 handcrafted GLCM texture descriptors for cataract grading
- Achieves 95% accuracy on 300 clinically labeled eye photos using only consumer-grade cameras
- Outperforms CNN-only (91.3%) and texture-only (88.5%) baselines without GPU acceleration
Why It Matters
Enables low-cost, scalable cataract screening in primary care and telemedicine, especially in resource-limited regions.