Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Skin Lesion Classification Using a Soft Voting Ensemble of Convolutional Neural Networks

Abdullah Al Shafi, Abdul Muntakim, Pintu Chandra Shill, Rowzatul Zannat, Abdullah Al-Amin

TL;DR

The paper tackles automated skin lesion classification by introducing a soft voting ensemble of CNNs trained on segmented dermoscopic images. Key innovations include a dual-encoder segmentation step to focus on lesion regions and a triad of pretrained backbones (MobileNetV2, VGG19, InceptionV3) whose predictions are fused via soft voting. Evaluations on HAM10000, ISIC 2016, and ISIC 2019 show high accuracies and strong recall, with ROC-AUC values suggesting robust discrimination across classes. The work emphasizes a balance between diagnostic accuracy and real-time applicability, aiming for practical deployment in clinical and mobile settings. Limitations remain in achieving optimal overall accuracy, motivating future work on improving reliability and efficiency.

Abstract

Skin cancer can be identified by dermoscopic examination and ocular inspection, but early detection significantly increases survival chances. Artificial intelligence (AI), using annotated skin images and Convolutional Neural Networks (CNNs), improves diagnostic accuracy. This paper presents an early skin cancer classification method using a soft voting ensemble of CNNs. In this investigation, three benchmark datasets, namely HAM10000, ISIC 2016, and ISIC 2019, were used. The process involved rebalancing, image augmentation, and filtering techniques, followed by a hybrid dual encoder for segmentation via transfer learning. Accurate segmentation focused classification models on clinically significant features, reducing background artifacts and improving accuracy. Classification was performed through an ensemble of MobileNetV2, VGG19, and InceptionV3, balancing accuracy and speed for real-world deployment. The method achieved lesion recognition accuracies of 96.32\%, 90.86\%, and 93.92\% for the three datasets. The system performance was evaluated using established skin lesion detection metrics, yielding impressive results.

Skin Lesion Classification Using a Soft Voting Ensemble of Convolutional Neural Networks

TL;DR

The paper tackles automated skin lesion classification by introducing a soft voting ensemble of CNNs trained on segmented dermoscopic images. Key innovations include a dual-encoder segmentation step to focus on lesion regions and a triad of pretrained backbones (MobileNetV2, VGG19, InceptionV3) whose predictions are fused via soft voting. Evaluations on HAM10000, ISIC 2016, and ISIC 2019 show high accuracies and strong recall, with ROC-AUC values suggesting robust discrimination across classes. The work emphasizes a balance between diagnostic accuracy and real-time applicability, aiming for practical deployment in clinical and mobile settings. Limitations remain in achieving optimal overall accuracy, motivating future work on improving reliability and efficiency.

Abstract

Skin cancer can be identified by dermoscopic examination and ocular inspection, but early detection significantly increases survival chances. Artificial intelligence (AI), using annotated skin images and Convolutional Neural Networks (CNNs), improves diagnostic accuracy. This paper presents an early skin cancer classification method using a soft voting ensemble of CNNs. In this investigation, three benchmark datasets, namely HAM10000, ISIC 2016, and ISIC 2019, were used. The process involved rebalancing, image augmentation, and filtering techniques, followed by a hybrid dual encoder for segmentation via transfer learning. Accurate segmentation focused classification models on clinically significant features, reducing background artifacts and improving accuracy. Classification was performed through an ensemble of MobileNetV2, VGG19, and InceptionV3, balancing accuracy and speed for real-world deployment. The method achieved lesion recognition accuracies of 96.32\%, 90.86\%, and 93.92\% for the three datasets. The system performance was evaluated using established skin lesion detection metrics, yielding impressive results.
Paper Structure (17 sections, 3 figures, 3 tables, 1 algorithm)

This paper contains 17 sections, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Proposed Methodology
  4. Dataset
  5. Preprocessing
  6. Segmentation
  7. Classification Models
  8. Pretrained models
  9. MobileNetV2
  10. VGG-19
  11. InceptionV3
  12. Proposed Ensemble Model
  13. Experimental Analysis and Results
  14. Experimental Setup
  15. Evaluation Metrics
  16. ...and 2 more sections

Figures (3)

  • Figure 1: The complete pipeline for segmentation and classification of skin lesion. The process starts with preprocessing steps, followed by segmentation using the dual encoder model. The segmented images are then passed to an ensemble of three pre-trained models (MobileNetV2, VGG19, InceptionV3) for classification.
  • Figure 5: Illustration of the dual encoder segmentation approach. The architecture consists of two encoders, VGG-16 and VGG-19, which are used in parallel to process the input image.
  • Figure 6: ROC curves for the classification performance of the proposed method on the (a) HAM10000, (b) ISIC 2019, and (c) ISIC 2016 datasets. The curves demonstrate the model's capacity to discriminate across classes, with AUC (Area Under Curve) values ranging from 0.87 to 0.96. Lower AUC of a class suggests reduced efficiency, while the higher AUC indicates reliable differentiation from the others.