Abstract:Skin cancer is the most common type of cancer. Specifically, melanoma is the cause of 75% of skin cancer deaths, although it is the least common skin cancer. Better detection of melanoma could have a positive impact on millions of people. The ISIC archive contains the largest publicly available collection of dermatoscopic images of skin lesions. In this research, we investigate the efficacy of applying advanced deep learning techniques in computer vision to identify melanoma in images of skin lesions. Through reviewing previous methods, including pre-trained models, deep-learning classifiers, transfer learning, etc., we demonstrate the applicability of the popular deep learning methods on critical clinical problems such as identifying melanoma. Finally, we proposed a processing flow with a validation AUC greater than 94% and a sensitivity greater than 90% on ISIC 2016 - 2020 datasets.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to use computer vision technology to improve the detection accuracy of melanoma (a type of skin cancer). Specifically, by analyzing skin lesion images in the ISIC 2016 - 2020 dataset, the researchers explored the effectiveness and applicability of various advanced deep - learning techniques in identifying melanoma. The goal of the paper is to develop a processing pipeline that can achieve an AUC (area under the curve) of over 94% and a sensitivity of over 90% on the validation set, thereby improving the early detection ability of melanoma in the clinical environment. ### Main Contributions 1. **Dataset Analysis**: The positive and negative sample distributions in the ISIC 2016 - 2020 dataset were discussed in detail, and how to alleviate the sample imbalance problem was studied. 2. **Model Architecture Experiments**: A large number of experiments were carried out to test the performance of different pre - trained models (such as AlexNet, VGG16, ResNet - 18, etc.) in the task. 3. **Technical Improvements**: Summarized the improvements of various techniques for the current task and proposed a processing pipeline that makes the validation AUC exceed 94% and the sensitivity exceed 90%. ### Method Overview 1. **Data Pre - processing**: - **Data Distribution**: Analyzed the ratio of positive and negative samples in the ISIC 2016 - 2020 dataset and found that the dataset was highly imbalanced. - **TFRecord Samples**: Converted the original JPEG images into TFRecord format for subsequent processing. 2. **Methodology**: - **Processing Pipeline**: Used K - fold cross - validation, input the enhanced images into the CNN model, extracted feature vectors, and finally classified them through an ensemble model. - **Image Enhancement**: Included operations such as random flipping, cropping, translation, rotation, and the Coarse Dropout technique. - **Oversampling**: Increased the number of positive samples through oversampling to alleviate the data imbalance problem. - **Backbone Networks**: Used multiple pre - trained models (such as Xception, VGG16, ResNet101 V2, etc.) for feature extraction. - **Evaluation**: Used AUC - ROC and sensitivity as performance evaluation indicators and adopted stratified K - fold cross - validation for data partitioning. 3. **Experimental Results**: - **Single - Model Experiments**: Tested the performance of multiple backbone networks under different input sizes and found that EfficientNet and VGG performed best in terms of validation AUC. - **Ensemble - Model Experiments**: Compared three ensemble learning strategies, namely Blending, MLP, and GBDT, and found that GBDT had the best effect, with a validation AUC exceeding 90%. - **Transfer - Learning Experiments**: Tried to use other related pre - trained models on the ISIC 2016 - 2020 dataset, but there was no significant performance improvement. - **Hyperparameter Optimization**: Used the Optuna tool for hyperparameter optimization to further improve the model performance. ### Conclusion This research, through detailed experiments and analysis, demonstrated the potential and effectiveness of deep - learning techniques in melanoma detection. The proposed processing pipeline achieved high AUC and sensitivity on the validation set, providing strong support for clinical applications.

A Comprehensive Evaluation Study on Risk Level Classification of Melanoma by Computer Vision on ISIC 2016-2020 Datasets

Intelligent Data Analytics for Diagnosing Melanoma Skin Lesions via Deep Learning in IoT System

Melanlysis: A Mobile Deep Learning Approach for Early Detection of Skin Cancer

The Application of Deep Learning in the Risk Grading of Skin Tumors for Patients Using Clinical Images

Enhancing Skin Cancer Diagnosis with Deep Learning-Based Classification

Melanoma Classification Using a Novel Deep Convolutional Neural Network with Dermoscopic Images

SIIM-ISIC Melanoma Classification With DenseNet

Assessing the effectiveness of artificial intelligence methods for melanoma:A retrospective review

Diagnosing Melanomas in Dermoscopy Images Using Deep Learning

Skin Cancer Classification Utilizing a Hybrid Model of Machine Learning Models Trained on Dermoscopic Images

Skin Cancer Classification and Comparison of Pre-trained Models Performance using Transfer Learning

Skin Lesion Analysis Towards Melanoma Detection Using Deep Learning Network

Melanoma identification and classification model based on fine-tuned convolutional neural network

Transforming Skin Cancer Diagnosis: A Deep Learning Approach with the Ham10000 Dataset

Detection and Classification of Melanoma Skin Cancer Using Image Processing Technique

A Comparative Analysis Towards Melanoma Classification Using Transfer Learning by Analyzing Dermoscopic Images

DePicT Melanoma Deep-CLASS: a deep convolutional neural networks approach to classify skin lesion images

AI outperformed every dermatologist: Improved dermoscopic melanoma diagnosis through customizing batch logic and loss function in an optimized Deep CNN architecture

Deep Learning Classifier with Patient's Metadata of Dermoscopic Images in Malignant Melanoma Detection

Melanoma Skin Cancer Detection Using Deep learning, Neural Network and Classical Machine Learning Techniques

On the Automatic Detection and Classification of Skin Cancer Using Deep Transfer Learning