Source-linked AI summary
Segmentation and Classification of Skin Lesions for Disease Diagnosis
Sumithra R, Mahamad Suhil, D. S. Guru
TL;DR
The paper addresses automatic skin-lesion diagnosis through segmentation and classification. It filters images, segments lesions with region growing, extracts color and texture features, and classifies them using SVM, k-NN, and their fusion. On a dataset of 726 samples from 141 images across five disease classes, SVM–k-NN fusion achieves the highest reported F-measure, while performance varies across classes.
Problem
Early skin-disease detection is more complex for inexperienced dermatologists, motivating computer-aided analysis of skin lesions.
Method
The system filters images, segments lesions with region growing, represents them using color and texture features, and applies SVM, k-NN, and decision fusion.
Results
61.03% F-measure is achieved by SVM–k-NN fusion, compared with 46.71% for SVM and 34.1% for k-NN using 70% training samples.
Takeaways & Limitations
The proposed method can be used as a supplementary tool for experts to diagnose skin diseases.
Takeaways & Limitations
Performance decreases considerably for some classes because the dataset was collected from internet resources.
Abstract
from arXiv · showhide
In this paper, a novel approach for automatic segmentation and classification of skin lesions is proposed. Initially, skin images are filtered to remove unwanted hairs and noise and then the segmentation process is carried out to extract lesion areas. For segmentation, a region growing method is applied by automatic initialization of seed points. The segmentation performance is measured with different well known measures and the results are appreciable. Subsequently, the extracted lesion areas are represented by color and texture features. SVM and k-NN classifiers are used along with their fusion for the classification using the extracted features. The performance of the system is tested on our own dataset of 726 samples from 141 images consisting of 5 different classes of diseases. The results are very promising with 46.71% and 34% of F-measure using SVM and k-NN classifier respectively and with 61% of F-measure for fusion of SVM and k-NN.
1. Introduction
Skin lesions are important clinical signs, but early diagnosis can be difficult for inexperienced dermatologists. This motivates computer-aided diagnosis using digital image processing and machine learning.
- Skin lesions can signal diseases such as chickenpox and melanoma.
- Early skin-disease detection is more complex for inexperienced dermatologists.
- Computer-aided diagnosis systems use digital image processing and machine learning to automate medical analysis.
2. Dataset Description
The paper constructs a five-class lesion dataset and processes images through filtering, segmentation, feature extraction, and classification. Lesions are represented using color, texture, and RGB histogram features.
- Dataset Description: 141 images cover five disease classes: melanoma, bullae, seborrheic keratosis, shingles, and squamous cell.The class counts are 31, 26, 33, 20, and 31 images, respectively.
- Dataset Description: 726 lesion samples are obtained by treating every lesion area after segmentation as a separate sample.
- Pre-processing and Segmentation: A 3×3 Gaussian filter smooths images to remove hair and noise before morphological erosion and dilation enhance them.
- Pre-processing and Segmentation: Region growing performs pixel-based segmentation after empirical initialization of seed points.The method aims to isolate lesion regions while limiting over- and under-segmentation.
- Color Features: 72 color features summarize four statistics across six color spaces and three channels.The statistics are mean, standard deviation, variation, and skewness.
- Feature Representation: Each lesion area is represented by 4182 features: 72 color, 14 texture, and 4096 RGB histogram features.Texture descriptors are based on gray level co-occurrence matrices, while RGB histograms use 16 bins per component sub-band.
4. Experimentation and Results
The experiments evaluate segmentation against dermatologist-marked ground truth and assess classification across training splits, classifier combinations, and disease classes. Fusion of SVM and k-NN achieves the strongest reported F-measure, while class-wise performance varies substantially.
- Segmentation Performance Analysis: Segmentation performance is compared with dermatologist-marked ground truth using overlap, under-segmentation, over-segmentation, similarity, and error measures.Higher overlap and similarity indicate better segmentation, whereas lower under-segmentation, over-segmentation, and error indicate better performance.
- Classification Performance Analysis: Classification experiments use random training/testing splits of 30:70, 40:60, 50:50, 60:40, and 70:30, with 20 repetitions per split.The k-NN value was selected empirically.
- Classification Performance Analysis: 61.03 F-measure is obtained by SVM and k-NN fusion with 70 percent of samples used for training, compared with 46.71 percent for SVM and 34.1 percent for k-NN.Both individual classifiers show high classification performance for this training proportion.
- Classification Performance Analysis: Class-wise F-measure is reported for SVM, k-NN, and their fusion using 70 percent of the samples for training.The results compare classifier behavior across individual disease classes.
- Classification Performance Analysis: Seborrheic keratosis achieves very good performance for all classifiers, whereas the remaining classes show poorer performance.The paper associates these differences with intra-class variation, interclass similarity, segmentation quality, and feature suitability.
- Classification Performance Analysis: A qualitative comparison with contemporary methods reports that the proposed method is superior in many respects.The comparison is presented in Table 5.1.
5. CONCLUSION
The proposed model segments and classifies skin lesions using region growing, color and texture features, and SVM, k-NN, or fused classification. It was evaluated across five disease classes, with fusion outperforming individual classifiers and the authors positioning the system as a supplementary diagnostic tool.
- Conclusion: Region growing segmentation, color and texture features, and SVM, k-NN, or fused classification comprise the proposed model.The model combines lesion segmentation with feature-based classification.
- Conclusion: The SVM-k-NN fusion classifier outperforms the individual classifiers.The paper reports this comparison as part of its classification performance evaluation.
- Conclusion: The authors conclude that the proposed method can serve as a supplementary tool for experts diagnosing skin diseases.This conclusion follows the reported classification results.
- Conclusion: The paper compares its proposed model with other contemporary skin-lesion segmentation and classification models.The comparison is described as a qualitative study of the proposed model against well-known models.
- Conclusion: Five disease classes are included in the authors’ dataset evaluation.The listed classes are Melanoma, Bullae, Seborrheic keratosis, Shingles, and Squamous cell.