Deep learning-based image segmentation for dermatological lesions
Generally, skin cancers, especially melanomas, have placed a huge health burden throughout the world, with the occurrence of more than 123,000 new cases annually. Early detection of melanoma is critical in preventing the progression of this disease into its invasive stages. However, most people are...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2025
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/6961/1/fyp_CN_2025_LJH.pdf http://eprints.utar.edu.my/6961/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1854094446172307456 |
|---|---|
| author | Lim, Jia Hong |
| author_facet | Lim, Jia Hong |
| author_sort | Lim, Jia Hong |
| building | UTAR Library |
| collection | Institutional Repository |
| content_provider | Universiti Tunku Abdul Rahman |
| content_source | UTAR Institutional Repository |
| continent | Asia |
| country | Malaysia |
| description | Generally, skin cancers, especially melanomas, have placed a huge health burden throughout the world, with the occurrence of more than 123,000 new cases annually. Early detection of melanoma is critical in preventing the progression of this disease into its invasive stages. However, most people are not giving enough attention to minor skin changes. Sometimes it is even difficult for doctors to distinguish between benign and malignant skin lesions. This study tries to solve it by proposing an automatic deep learning system for segmentation and classification in skin lesion images. This paper proposes a system that incorporates the use of a U-Net-based CNN and DeepLabV3, which proves helpful in the segmentation of an image in such a way that accurate mask. Furthermore, with the helping of the classifier model (ResNet-18) to classify the moles condition such as benign or malignant. The implementation of the system will be foreseen to enhance the diagnostic process, minimizing the time and difficulty brought forth by current methods, including invasive procedures and waiting for test results. This system uses a huge dermatological image database in order to apply deep learning methodologies for classifying a skin lesion with high precision. It will also be able to separate melanomas into either benign or malignant. The proposed automated system can lead to early diagnosis of the disease, which helps in effective early treatment, hence reducing the skin cancer burden. Moreover, the developed system will allow for patients and physicians to upload images through user-friendly web interface showing immediate real-time analysis and diagnosis. This is intended to run smoothly on the web interface, making it accessible both in clinical settings and possibly integrated into web interface for remote diagnostics. The research is done to show how deep learning can radically enhance the speed and accuracy of skin cancer diagnosis through segmentation and classification of skin lesions. If this can be affected with not much loss of time, it could save many lives by early detection and intervention. |
| format | Final Year Project / Dissertation / Thesis |
| id | my-utar-eprints.6961 |
| institution | Universiti Tunku Abdul Rahman |
| publishDate | 2025 |
| record_format | eprints |
| spelling | my-utar-eprints.69612025-12-28T10:45:10Z Deep learning-based image segmentation for dermatological lesions Lim, Jia Hong T Technology (General) TD Environmental technology. Sanitary engineering Generally, skin cancers, especially melanomas, have placed a huge health burden throughout the world, with the occurrence of more than 123,000 new cases annually. Early detection of melanoma is critical in preventing the progression of this disease into its invasive stages. However, most people are not giving enough attention to minor skin changes. Sometimes it is even difficult for doctors to distinguish between benign and malignant skin lesions. This study tries to solve it by proposing an automatic deep learning system for segmentation and classification in skin lesion images. This paper proposes a system that incorporates the use of a U-Net-based CNN and DeepLabV3, which proves helpful in the segmentation of an image in such a way that accurate mask. Furthermore, with the helping of the classifier model (ResNet-18) to classify the moles condition such as benign or malignant. The implementation of the system will be foreseen to enhance the diagnostic process, minimizing the time and difficulty brought forth by current methods, including invasive procedures and waiting for test results. This system uses a huge dermatological image database in order to apply deep learning methodologies for classifying a skin lesion with high precision. It will also be able to separate melanomas into either benign or malignant. The proposed automated system can lead to early diagnosis of the disease, which helps in effective early treatment, hence reducing the skin cancer burden. Moreover, the developed system will allow for patients and physicians to upload images through user-friendly web interface showing immediate real-time analysis and diagnosis. This is intended to run smoothly on the web interface, making it accessible both in clinical settings and possibly integrated into web interface for remote diagnostics. The research is done to show how deep learning can radically enhance the speed and accuracy of skin cancer diagnosis through segmentation and classification of skin lesions. If this can be affected with not much loss of time, it could save many lives by early detection and intervention. 2025-06 Final Year Project / Dissertation / Thesis NonPeerReviewed application/pdf http://eprints.utar.edu.my/6961/1/fyp_CN_2025_LJH.pdf Lim, Jia Hong (2025) Deep learning-based image segmentation for dermatological lesions. Final Year Project, UTAR. http://eprints.utar.edu.my/6961/ |
| spellingShingle | T Technology (General) TD Environmental technology. Sanitary engineering Lim, Jia Hong Deep learning-based image segmentation for dermatological lesions |
| title | Deep learning-based image segmentation for dermatological lesions |
| title_full | Deep learning-based image segmentation for dermatological lesions |
| title_fullStr | Deep learning-based image segmentation for dermatological lesions |
| title_full_unstemmed | Deep learning-based image segmentation for dermatological lesions |
| title_short | Deep learning-based image segmentation for dermatological lesions |
| title_sort | deep learning-based image segmentation for dermatological lesions |
| topic | T Technology (General) TD Environmental technology. Sanitary engineering |
| url | http://eprints.utar.edu.my/6961/1/fyp_CN_2025_LJH.pdf http://eprints.utar.edu.my/6961/ |
| url_provider | http://eprints.utar.edu.my |