Artificial Intelligence for localization and segmentation of organs in medical imaging

Monday 1 July 2024, 10:30 am
PhD candidate
G.E. Humpire Mamani
Promotor(s)
prof. dr. B. van Ginneken, prof. dr. W.M. Prokop
Co-promotor(s)
dr. ir. C. Jacobs, dr. N. Lessmann
Location
Aula

This thesis explores the use of deep learning to enhance the detection and segmentation of structures in CT scans, particularly for oncology patients. Chapter 2 introduces a method for localizing organs in 3D using orthogonal views. Chapter 3 details a CNN-based segmentation algorithm for the spleen, showing high clinical readiness. Chapter 4 addresses the segmentation of kidney abnormalities, highlighting an ablation study to optimize performance. Chapter 5 demonstrates the use of transfer learning to segment additional structures from a partially annotated dataset. The work emphasizes the potential of automated tools to reduce radiologists' workloads, improve accuracy, and aid in clinical decision-making by providing reliable segmentation and localization, which are crucial for diagnosis and treatment planning. The thesis concludes that these methods can significantly contribute to computer-aided detection and diagnosis systems.

Gabriel E Humpire-Mamani graduated from Systems Engineering at the National University of San Agustin, Peru. In 2012, he obtained a master’s degree in computer science at the University of Sao Paulo, Brazil. In February 2016, he started working as a Ph.D. student at DIAG. His doctoral research focused on detecting and segmenting organs in cancer patients.