Bi-directional Dermoscopic Feature Learning and Multi-scale Consistent Decision Fusion for Skin Lesion Segmentation
Summary
This study introduces a new framework for segmenting skin lesions in dermoscopic images, improving melanoma diagnosis. The bi-directional dermoscopic feature learning (biDFL) method enhances accuracy by modeling lesion context effectively.
Area of Science:
- Medical Imaging
- Computer-Aided Diagnosis
- Dermatology
Background:
- Accurate skin lesion segmentation is vital for melanoma diagnosis using computer-aided methods.
- Dermoscopic image variations present challenges in anatomical learning and consistent lesion delineation.
Purpose of the Study:
- To propose a novel bi-directional dermoscopic feature learning (biDFL) framework.
- To enhance the modeling of complex correlations between skin lesions and their context.
- To improve the accuracy and consistency of skin lesion segmentation.
Main Methods:
- Developed a bi-directional dermoscopic feature learning (biDFL) framework.
- Integrated biDFL modules into a Convolutional Neural Network (CNN) architecture.
- Introduced a multi-scale consistent decision fusion (mCDF) mechanism for improved delineation.
Main Results:
- Achieved a substantially rich and discriminative feature representation.
- Enhanced high-level parsing performance through biDFL modules.
- Demonstrated state-of-the-art performance in skin lesion segmentation on public datasets.
Conclusions:
- The proposed biDFL framework effectively models lesion context for improved segmentation.
- The mCDF method enhances delineation by analyzing decision consistency across multiple layers.
- The method shows significant effectiveness and consistency on multiple dermoscopic image databases.
