Software-Engineering Perspectives on Machine for Skin-Disease Classification

Skin‑disease classification has evolved from simple image recognizers into software‑driven pipelines that demand reliability, reproducibility, and ethical governance. While most AI reviews focus on algorithmic accuracy, few examine these systems through a software‑engineering (SE) lens—essential for assessing pipeline modularity, version control, deployment readiness, and long‑term maintainability, all critical for clinical integration. This review surveys literature from 2015 to early 2025, curating about 180 papers that link skin‑disease classification with SE practices. It traces the shift from handcrafted feature‑based classifiers to end‑to‑end convolutional, ensemble, and transformer architectures, alongside the engineering processes that support versioning, deployment, and monitoring. Benchmark datasets (PH$^2$, HAM10000, ISIC, etc.) have established reproducible evaluation protocols that underpin software verification. Emerging directions—self‑supervised pretraining, multimodal fusion, human‑AI collaboration—signal a move from model‑centric to system‑level integration. The analysis highlights not only accuracy and generalization but also SE quality attributes: scalability, maintainability, explainability, and fairness, which are indispensable for trustworthy adoption in diverse clinical workflows.