| _version_ | 1866901156336238592 |
|---|---|
| author | Malhotra, Siddharth |
| author_facet | Malhotra, Siddharth |
| contents | <p>This study presents a deep learning–based framework for the automated classification of skin cancer using the HAM10000 dataset, which contains over 10,000 dermatoscopic images across seven lesion categories. Early detection of skin cancer is critical for improving patient survival rates, and computer-aided diagnostic systems can play a vital role in supporting clinicians.</p> <p>To address dataset imbalance and improve generalization, images were preprocessed through scaling, normalization, and extensive augmentation techniques. The proposed model leverages transfer learning with the Xception architecture, where the final 50 layers were fine-tuned to capture domain-specific features. A classification head incorporating dense layers, dropout, and batch normalization was applied to reduce overfitting.</p> <p>The system achieved strong performance, with a recall of 0.9920, precision of 0.9933, accuracy of 99.25%, and a categorical cross-entropy loss of 0.0283. Comparative results indicate that the Xception-based approach significantly outperforms conventional CNN models, which typically achieve around 85% accuracy.</p> <p>This work highlights the potential of advanced deep learning methods for reliable skin cancer diagnosis and provides a reproducible framework that can be extended to other medical imaging tasks.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17227860 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Deep Learning Models for Medical Imaging in Skin Cancer Diagnosis Malhotra, Siddharth <p>This study presents a deep learning–based framework for the automated classification of skin cancer using the HAM10000 dataset, which contains over 10,000 dermatoscopic images across seven lesion categories. Early detection of skin cancer is critical for improving patient survival rates, and computer-aided diagnostic systems can play a vital role in supporting clinicians.</p> <p>To address dataset imbalance and improve generalization, images were preprocessed through scaling, normalization, and extensive augmentation techniques. The proposed model leverages transfer learning with the Xception architecture, where the final 50 layers were fine-tuned to capture domain-specific features. A classification head incorporating dense layers, dropout, and batch normalization was applied to reduce overfitting.</p> <p>The system achieved strong performance, with a recall of 0.9920, precision of 0.9933, accuracy of 99.25%, and a categorical cross-entropy loss of 0.0283. Comparative results indicate that the Xception-based approach significantly outperforms conventional CNN models, which typically achieve around 85% accuracy.</p> <p>This work highlights the potential of advanced deep learning methods for reliable skin cancer diagnosis and provides a reproducible framework that can be extended to other medical imaging tasks.</p> |
| title | Deep Learning Models for Medical Imaging in Skin Cancer Diagnosis |
| url | https://doi.org/10.5281/zenodo.17227860 |