University of Twente Student Theses
Symmetry enhanced skin lesion classification network : a cascaded multi-task learning approach
Ramakrishnan, Akash (2024) Symmetry enhanced skin lesion classification network : a cascaded multi-task learning approach.
|
PDF
6MB |
| Abstract: | Skin lesions represent a category of dermatological conditions where timely and accurate analysis is crucial for preventing malignancy. Assessing the symmetry of a lesion is a critical factor in determining its malignancy. This report presents a Multi-Task Learning (MTL) approach to skin lesion analysis that leverages lesion symmetry as a key feature for skin lesion classification. The study proposes a cascaded MTL architecture for comprehensive skin lesion analysis, incorporating three distinct tasks: skin lesion segmentation, lesion symmetry classification, and skin lesion classification. The proposed Symmetry Enhanced Lesion Classification Network (SE-LCN) leverages segmentation masks to refine the performance of two classification networks, enhancing the overall diagnostic accuracy. Class Activation Maps (CAMs) from the symmetry classification network are employed to augment the lesion classification network, aiming to provide more targeted and effective predictions. An extensive ablation study is conducted to analyze the impact of using various masking strategies on both classification networks and the efficiency of CAM transfer between the networks. This study not only analyzes the impact of these features on the predictive performance of these networks, but also the localization accuracy of the CAMs generated by these networks. The results demonstrate that integrating CAMs improves the predictive performance, and localization accuracy of the lesion classification network, validating the effectiveness of the proposed cascaded MTL architecture. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Programme: | Electrical Engineering MSc (60353) |
| Link to this item: | https://purl.utwente.nl/essays/102063 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page