Cascade Classification of Dermoscopic Images of Skin Neoplasms with Controllable Sensitivity and External Clinical Validation

2026-06-11 · Source: Artificial Intelligence · Field: Science & Research — Health & Medical Research, Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Advanced, quick

Summary

The study compared four deep learning architectures (ViT-B/16, Swin-S, ConvNeXt-S, EfficientNetV2-S) and three classification schemes for dermoscopic images of skin neoplasms. Models were trained on aggregated ISIC Archive data using ImageNet-pretrained weights and evaluated on internal and two independent clinical datasets from Russian practice (Melanoscope AI, Sechenov University). Internally, the binary classification stage achieved ROC-AUC 0.952-0.966, but on Sechenov University data, ROC-AUC dropped to 0.797-0.893, sensitivity to 0.53-0.67, and ECE increased from 0.02 to 0.27-0.39, indicating a significant generalization gap. The two-stage cascade classification scheme improved macro F1 over single-stage four-class classification, particularly for ViT-B/16, by recovering malignant lesions.

Key takeaway

For AI Scientists developing diagnostic tools for dermoscopic images, recognize that a significant generalization gap exists between open datasets and real-world clinical data. You should prioritize implementing cascade classification schemes with tunable triage thresholds to achieve controllable sensitivity and ensure robust external clinical validation and recalibration before any deployment. This approach better aligns with clinical differential-diagnosis logic and improves overall performance.

Key insights

Cascade classification with a tunable triage threshold offers superior sensitivity control and better mimics clinical diagnosis logic.

Principles

• External clinical validation is crucial for generalization.
• Deep learning models show a significant generalization gap.
• Cascade classification can recover misclassified lesions.

Method

The study compared binary, single-stage four-class, and two-stage cascade classification schemes using ViT-B/16, Swin-S, ConvNeXt-S, and EfficientNetV2-S architectures on dermoscopic images.

In practice

• Implement a two-stage cascade for skin lesion classification.
• Prioritize external validation for model deployment.
• Consider ViT-B/16 with cascade for improved F1.

Topics

• Dermoscopic Image Classification
• Cascade Classification
• Deep Learning Architectures
• Clinical Validation
• Generalization Gap
• Skin Neoplasms

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Clinically-Informed Modeling for Pediatric Brain Tumor Classification from Whole-Slide Histopathology Images — Contrastive learning, especially with expert-guided hard negatives, improves fine-grained pediatric brain tumor classification from WSIs.
• Validating a Deep Learning Algorithm to Identify Patients with Glaucoma using Systemic Electronic Health Records — An EHR-only deep learning model can effectively pre-screen for glaucoma without specialized imaging.
• CASCADE Conformal Prediction: Uncertainty-Adaptive Prediction Intervals for Two-Stage Clinical Decision Support — The framework uses epistemic uncertainty from a classifier to adaptively scale prediction intervals for a regression task, improving clinical decision support.
• DaX: Learning General Pathology Representations Across Scales — DaX is a pathology foundation model using DINOv3-style self-supervised learning to create robust, transferable visual representations across diverse scales and clinical tasks.
• Computational Imaging Priors for Wireless Capsule Endoscopy: Monte Carlo-Guided Hemoglobin Mapping for Rare-Anomaly Detection — A Monte Carlo-inspired prior improves wireless capsule endoscopy classification by better distinguishing hemoglobin from confounding factors.
• Bridging the interpretability gap for medical artificial intelligence models using class-association manifold learning — CAML enhances medical AI explainability by decoupling decision patterns and visualizing global class-associated knowledge.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.