Metastatic bone disease arises when a primary cancer has spread to the bone. Quantitative information about the metastatic lesions can be assessed from inspecting medical images such as MRI. Metastatic bone lesions are often spread across the whole body, and are small and irregular in shape making manual assessment labour intensive and prone to errors. Deep learning has the potential to guide medical doctors in their decision making by automatically extracting quantitative, reproducible features from the medical images.
Compared to classical computer vision, medical images come with different challenges. The most prominent difference is that 3D whole-body images are larger in size compared to even the highest resolution pictures while regions of interest are often relatively small. To reduce the need of expensive computational resources, deep learning on medical images is often performed only on patches rather than full images. This comes at the cost of global information that can be leveraged by the models such as the location of the patch inside of the body. Many diseases, such as metastatic bone disease, don’t spread homogeneously in the body. Metastatic bone disease occurs more frequently in the pelvic region compared to more distal bones in the body. The location in the body thus yields relevant information for disease assessment.
The aim of this PhD is to develop a novel location aware deep learning pipeline for whole-body segmentation of metastatic bone disease on MRI. Location embeddings that yield anatomical information of the location with respect to the body will be developed and added to the models. The developed system will provide accurate quantitative information about the disease involvement of the metastatic bone lesions.
Joris obtained his bachelor of civil engineering in 2017 at the Ugent. In 2020, he obtained the master of science degree in Biomedical engineering. Continuing his research at ETRO that started during his master thesis, Joris joined the ETRO-MIT team as an academic researcher from october 2020. Research interests: Computer vision, Deep learning, Medical image processing, computer aided diagnosis.