Work

Development of advanced machine learning models for the fusion of heterogeneous biological sources in clinical decision support systems for cancer

Year: 2023
Type: dissertation
Abstract: Cancer is one of the leading causes of death worldwide, just behind cardiovascular diseases. \nAn early diagnosis is key for the prognosis of the patient, since it allows applying the most \nsuitable treatment. To do so, multiple screenings are routinely performed on the patient \ninvolving, for instance, the visual examination of histopathological slides, the analysis of the \nclinical history, or finding alterations in their gene expression. These examinations, however, \nare usually time-consuming, and not always the physicians have the experience to analyze \nthem. To help them with these tasks, clinical decision support systems have been created in \nrecent years using the advances in the machine learning field. Machine learning models \nare able to automatically learn from these data, and find insights that can help them to \nsolve a specific task. This is part of the precision medicine field where, using a data-driven \napproach, we tailor the diagnosis, treatment, and other clinical outcomes to the specific \ncharacteristics of the patient. Thanks to the advances in this field, more heterogeneous \nsources of biological information are being gathered, and they provide diverse features that \ncan help to accurately diagnose a cancer patient. This allows to create systems that use all \nthe available information, accurately modelling the patient’s disease. This would be similar \nto having a separate diagnosis per data modality from a group of expert clinicians, where \nthe final diagnosis is based on their analysis of their source of expertise. Unfortunately, \nnot all these sources are always available, limiting the potential of creating multi-modal \nmachine learning models. \nIn this thesis, we explore the improvements that can be obtained by using multi-modal \nmachine learning models resilient to missing modalities over single-modality ones in \nthe area of cancer diagnosis. Firstly, we tackled the problem of lung cancer subtyping \ndiagnosis using two of the most-used biomedical modalities in literature (gene expression \nand histopathology images), showing the improvements that can be obtained by fusing \nthese two modalities in comparison to being independently used. Next, to study the limits \nthat can be achieved by fusing heterogeneous biological sources, we include three new \nmodalities to the proposed problem (micro-RNA, DNA Methylation values, and the copy \nnumber variation of the genes). We tested which modalities complemented each other, \nand which is the performance that can be obtained by fusing all these modalities in a \nclassification model. Lastly, we approached the problem of data scarcity in biomedical \nmulti-modal problems, presenting advance methodologies for biological data generation. \nInspired by the recent advances in multi-modal generative models for natural images, we \nfocus on generating one modality based on a paired one (RNA-to-image synthesis problem) \nfor healthy tissues.We showed how the synthetic generated data were similar to the real \nsamples and the model was able to impute missing modalities.
Language: English
Open Access status: green
OpenAlex ID: W7008969353
Cites: 227
Cited by: 1
Related to: 10