shahidzikria/ADD-Net

ADD-Net is a typical academic/research artifact with very low community traction (16 stars over ~3.6 years). While it claims superior performance (98.6% accuracy) against standard models like DenseNet and VGG, these results are obtained on public Kaggle datasets which are known for data leakage and do not translate to clinical settings. The 'moat' in medical AI is not the specific CNN architecture—which is easily reproducible—but rather clinical validation, FDA/CE clearance, and access to proprietary, longitudinal patient data. Frontier labs are unlikely to target this specific niche directly, but the project is effectively displaced by modern medical imaging foundation models and vision transformers (ViTs) that have emerged since this project's last update. There is no evidence of active maintenance or an ecosystem, making it a static reference implementation rather than a viable tool.