TAZ-TFM-2024-735

Comparative visual analysis for multi-modal liver data

Hernández Alemán, Hugo
Raidou, Renata Georgia (Supervisora en la Universidad Técnica de Viena) (dir.) ; Tellería Oriols, Carlos (dir.)

Universidad de Zaragoza, EINA, 2024
Informática e Ingeniería de Sistemas department, Lenguajes y Sistemas Informáticos area

Máster Universitario en Ingeniería Biomédica

Abstract: Radiomics refers to the field of study that involves the extraction and analysis of a large number of quantitative features from medical imaging data, allowing for a more comprehensive and quantitative understanding of diseases. Radiomics is particularly useful in oncology for cancer diagnosis, treatment planning, and monitoring. On the other hand, Visual Analytics (VA) is an interdisciplinary field that combines interactive visualizations with analytical techniques to explore and understand complex datasets. The main goals are to facilitate the exploration of data, to enable users to gain insights (patterns, outliers, meaningful information...etc), and to make informed decisions. By bringing together the benefits of both radiomics and VA, researchers and clinicians can get important information improving the understanding and management of diseases. In this work, we present the development of a VA framework for radiomic data analysis, providing researchers and clinicians with an intuitive platform to extract meaningful insights from the medical images of liver cancer patients. This cohort comprises a total of 19 individuals sourced from the University of San Diego, some of whom have Hepatocellular Carcinoma (HCC) or Metastasis (MET). Furthermore, these patients have undergone radiotherapy, and their data also includes information on whether they have experienced any adverse effects (toxicity or non-toxicity). Imaging studies in the context of radiotherapy often focus on various aspects related to the treatment process, so in our dataset, we are able to study the breathing cycle, treatment average, and dose distribution. Our framework uses all these radiomic features to characterize tumors, identify patterns or correlations, and make cohort stratification. As a trial to unveil new information, Couinaud segmentation of the liver is performed, useful in radiology for planning and helping in localizing specific radiomic features of liver segments. We evaluate the stratification of patients using certain metrics such as the Silhouette Score, Inertia, or the Davies-Bouldin index. Additionally, by utilizing radiomic data and dosage information, we try to find differences between the patient groups we have (hepatocarcinoma/metastatic or toxicity/non-toxicity) through statistical tests or graphical representations, and also the prediction of toxicity. Finally, knowing the precise position of the tumor thanks to Couinaud's segments and dosage measurements, we identify areas of the liver that may be over or under-irradiated.

Universidad de Zaragoza Repository

+

-