Machine learning based abdominal multi-organ segmentation

Surgery is the oldest type of cancer therapy and remains an effective treatment for many types of cancer today. Surgery is often challenging, as the tumor is not always easy to distinguish from healthy tissue and surrounding healthy tissue should be spared to limit morbidity. Intraoperative surgical navigation is a new technique that is a part of the operating room today at the Netherlands Cancer Institute (NKI) which is promoting safer, more tissue sparing and accurate oncological surgical procedures.

For proper surgery planning, it is a must to have a preoperative comprehensive understanding of the internal anatomy of the patient. Both computer tomography (CT) and magnetic resonance imaging (MRI) play an important role in the visualization of the anatomy. To improve the understanding of these images, segmentation of the organs in the abdomen region can be used to build a 3D anatomical model. This model is subsequently input for the navigation system, which is used to guide the surgeon during the operation.

Segmentation is a cumbersome task due to the need of annotating hundreds of imaging slices and perform those annotations while considering a continuous and smooth labeling. Furthermore, manual segmentation is known to be variable between observers. Having an automatic and accurate segmentation of the abdominal organs will speed up the planning process and will improve the accuracy of navigation.

Machine learning regained a lot of popularity in the last few years especially in the medical image analysis field. Advancement in hardware development lead to stronger GPUs with greater memory. Availability of efficient software libraries again opened a door for gaining popularity, especially the convolutional neural networks have seen an increase in interest due to their highly computational requirements. This master 3 internship aims to apply and potentially improve state of the art algorithms for automatic segmentation across multi modal imaging techniques. The abundance of data present in the NKI allows for multi modal comparison and evaluation. Possibilities of exploring segmentation challenges span multiple abdominal organs such as the prostate and liver.

Prerequisites

• Enthusiastic Master student in electrical engineering, biomedical engineering, computer science, technical medicine or a related field
• A good team player with excellent communication skills
• A creative solution-finder

Duration: 10 weeks (M2) or 40 weeks (M3)

Start date: a.s.a.p.