Multimodal Data Fusion and Machine Learning for Advancing Biomedical Applications

This dissertation delves into the intricate landscape of biomedical imaging, examining the transformative potential of data fusion techniques to refine our understanding and diagnosis of health conditions. Daily influxes of diverse biomedical data prompt an exploration into the challenges arising from relying solely on individual imaging modalities. The central premise revolves around the imperative to combine information from varied sources to achieve a holistic comprehension of complex health issues.

The chapters included here contain articles and excerpts from published works. The study unfolds through an examination of four distinct applications of data fusion techniques. It commences with refining clinical task performance in chest X-ray classification and progresses to predicting prostate cancer by fusing whole-slide optical data with mass spectrometry imaging. The research extends further to the intricate task of reconstructing high-resolution models of a mouse brain tissue sample and optimizing image resolution in Raman spectroscopy for scrutinizing chemical substances.

While showcasing the efficacy of data fusion in diverse biomedical scenarios, the dissertation also addresses associated challenges. These challenges include data heterogeneity, scarcity, ensuring quality and consistency across multiple datasets, and complexities in aligning and registering data from disparate modalities. The dissertation underscores the pivotal role of data fusion in advancing biomedical imaging practices, providing nuanced insights into how integrating information from varied imaging modalities can significantly enhance diagnostic precision. In essence, this research contributes valuable insights and fosters advancements in the understanding and treatment of health issues, with a particular focus on the potential and challenges associated with data fusion techniques in biomedical imaging.