CI-Net

CI-Net performs automated skin lesion recognition from dermoscopy images to support skin cancer diagnosis by simulating dermatologists' diagnostic steps.

Key Features:

• Medical Knowledge Integration: Incorporates clinical diagnostic knowledge to align analysis with practical strategies used by dermatologists.
• Simulation of Diagnostic Process: Implements a three-step diagnostic strategy (zoom, observe, compare) mapped to specific computational modules for lesion assessment.
• Lesion Area Attention Module: Focuses on identifying and isolating the region of interest within dermoscopy images.
• Feature Extraction Module: Extracts detailed lesion characteristics for downstream analysis.
• Lesion Feature Attention Module: Emphasizes distinguishing features to support differentiation between lesion types.
• Distinguish Module: Implements decision-making strategies to differentiate benign and malignant skin conditions.

Scientific Applications:

• Dermatology — Automated skin cancer diagnosis: Applied to automated skin lesion classification from dermoscopy images and evaluated on ISIC 2016, ISIC 2017, ISIC 2018, ISIC 2019, ISIC 2020, and PH2 datasets with reported improvements in accuracy over existing methods.

Methodology:

The method comprises specialized computational modules: a lesion area attention module for lesion localization, a feature extraction module for detailed feature representation, a lesion feature attention module for highlighting discriminative features, and a distinguish module that implements decision-making strategies to differentiate skin conditions.