Implementation of personalized medicine in malignant melanoma patients aided by artificial intelligence
(English) Skin cancer, characterised by the uncontrolled growth of abnormal skin cells, poses a significant global health challenge. Melanoma, its most aggressive form, is particularly concerning due to its rapid metastasis if not detected early, leading to a sharp decline in survival probability. T...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2024 |
| País: | España |
| Institución: | CBUC, CESCA |
| Repositorio: | TDR. Tesis Doctorales en Red |
| OAI Identifier: | oai:www.tdx.cat:10803/692510 |
| Acceso en línea: | http://hdl.handle.net/10803/692510 https://dx.doi.org/10.5821/dissertation-2117-417796 |
| Access Level: | acceso abierto |
| Palabra clave: | Àrees temàtiques de la UPC::Enginyeria biomèdica Àrees temàtiques de la UPC::Informàtica 004 616 |
| Sumario: | (English) Skin cancer, characterised by the uncontrolled growth of abnormal skin cells, poses a significant global health challenge. Melanoma, its most aggressive form, is particularly concerning due to its rapid metastasis if not detected early, leading to a sharp decline in survival probability. This underscores the critical importance of timely diagnosis and treatment. While advances have been made, the vast surface area of the skin makes early detection challenging. The emergence of prospective clinical and imaging databases has revolutionised the field, providing sensitive and specific biomarkers for non-invasive cancer diagnosis. This thesis harnesses these advancements, introducing AI-powered tools for melanoma analysis using clinical data and Whole Slide Imaging (WSI). The focus lies on identifying early-stage melanoma through risk grouping and biomarker detection. By employing advanced survival analysis, pattern recognition, and statistical clustering, the research develops predictive and interpretable models to enhance early detection and diagnosis. In our study, we comprehensively evaluate survival analysis algorithms on melanoma datasets, highlighting the superior performance of tree-based methods over deep learning models in this context. We also detail the development of SurvLIMEpy, an open-source Python library for model explainability in survival analysis, fostering trust between clinicians and AI. This library has garnered significant attention, with over 10,000 downloads. Furthermore, our research demonstrates that machine learning models prioritise clinically relevant features for survival predictions, further reinforcing their interpretability and clinical utility. We also showcase the successful application of machine learning for patient stratification, outperforming the AJCC staging system used by dermatologist and enabling more personalised treatment strategies. Additionally, the thesis explores AI-driven biomarker prediction from WSIs underscores this approach's promise while emphasising the need for larger datasets for clinical implementation. This research signifies an advancement in the application of AI for melanoma analysis. The SurvLIMEpy library has empowered the research community, while the findings on patient stratification and biomarker prediction potentially improve melanoma diagnosis and treatment. |
|---|