Related Experiment Video
Updated: Jul 23, 2025
03:47
Author Spotlight: Advancing Facial Rejuvenation Therapy with Post-Laser Salicylic Acid Application
Published on: September 27, 2024
930
Refining skin lesions classification performance using geometric features of superpixels
Simona Moldovanu1,2, Mihaela Miron1, Cristinel-Gabriel Rusu2,3
1Department of Computer Science and Information Technology, Faculty of Automation, Computers, Electrical Engineering and Electronics, Dunarea de Jos University of Galati, 47 Domneasca Str., 800008, Galati, Romania.
Scientific Reports
|July 15, 2023
Summary
This study enhances skin lesion detection using improved Simple Linear Iterative Clustering (iSLIC) superpixels for accurate melanoma and nevus classification. The novel approach improves diagnostic accuracy in dermoscopy images.
Area of Science:
- Medical Image Analysis
- Computer Vision
- Computational Dermatology
Background:
- Accurate detection and classification of skin lesions, such as melanoma and nevi, are critical for early diagnosis and treatment.
- Dermoscopy images provide detailed views of skin lesions, but automated analysis remains challenging.
- Existing methods often struggle with precise segmentation and classification, leading to potential diagnostic errors.
Purpose of the Study:
- To introduce an improved Simple Linear Iterative Clustering (iSLIC) superpixels algorithm for enhanced skin lesion segmentation in dermoscopy images.
- To develop a robust method for discriminating between melanoma and nevi without false negatives.
- To evaluate the performance of machine learning and neural network classifiers using extracted features from segmented superpixels.
Main Methods:
- Proposed an improved Simple Linear Iterative Clustering (iSLIC) algorithm for superpixel generation and image segmentation.
- Utilized a local graph cut method to identify and isolate regions of interest (skin lesions).
- Extracted shape and geometric features from segmented superpixels and fed them into various machine learning (Random Forest, SVM, AdaBoost, KNN, DT, GNB) and neural network (PRNN, FFNN, 1D-CNN) classifiers.
Main Results:
- The iSLIC algorithm effectively segmented skin lesions, discarding background superpixels.
- The proposed method achieved high accuracy in classifying skin lesions, outperforming existing state-of-the-art methods.
- Evaluations on the 7-Point MED-NODE and PAD-UFES-20 datasets demonstrated the superior performance of the developed approach.
Conclusions:
- The integration of iSLIC superpixels and advanced machine learning/neural network models offers a powerful tool for accurate skin lesion detection and classification.
- This method shows significant potential for improving the diagnostic accuracy of melanoma and nevus identification in clinical settings.
- The proposed approach successfully addresses the challenge of false negatives in skin lesion classification.
