ISCAP – DM - Business Intelligence and Analytics
Permanent URI for this collection
Browse
Browsing ISCAP – DM - Business Intelligence and Analytics by Author "Casais, João Pedro Miranda"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Multimodal data integration in oncology: a case study of LIHCPublication . Casais, João Pedro Miranda; Ramos, Patrícia Alexandra Gregório; Oliveira, José Manuel SoaresThis thesis focuses on the application of multimodal learning, specifically deep learning techniques, to enhance the diagnosis and treatment of Liver Hepatocellular Carcinoma. The increasing incidence and late detection of this aggressive form of cancer necessitate innovative approaches for early identification and personalized therapies. The core objective of this thesis was to evaluate the impact of integrating diverse data modalities, such as clinical, genetic, and imaging data, on improving diagnostic precision and therapeutic outcomes for Liver Hepatocellular Carcinoma patients. The study investigates the limitations of traditional unimodal data analysis in understanding the complex nature of Liver Hepatocellular Carcinoma. It posits that multimodal learning can leverage complementary strengths of different data types, leading to a more comprehensive understanding of tumor biology and individual patient profiles. The research explores the use of advanced deep learning architectures for processing and integrating these diverse data modalities. A key component of the thesis involves a practical application using a dataset sourced from the American National Cancer Institute, focusing on predicting the vital status of Liver Hepatocellular Carcinoma patients. The dataset is preprocessed and analyzed using the AutoGluon-Multimodal (AutoMM) framework, an open-source automated machine learning library designed for multimodal learning tasks. Various model combinations incorporating image, text, and tabular data are evaluated based on classification performance metrics, including accuracy, Matthews Correlation Coefficient, precision, recall, F1 score, and ROC-AUC. The results demonstrate the superior performance of models combining tabular and text data in predicting the vital status of Liver Hepatocellular Carcinoma patients.