Liver tumor project

Challenge

We carried out a project for one of the leading companies in the pharmaceutical industry. The client had labeled CT scans of a liver with HCC-type cancer. They required a model that would segment liver tumors and provide objective data in the form of volumes of these tumors. A radiomics component was to be included in the design.

What we did

First of all, the model had to be created from scratch. Our goal was to get the best possible results. To train the model we used CT scans provided by the client. In the project we used elements of radiomic analysis, which consists in the fact that the algorithm looks for different features in a given organ and groups them (shape, structure/texture, signal intensity) that cannot be identified by a human. It then looks for relationships between them. The relationship is sometimes so subtle that a human is unable to find such non-obvious connections.

Results

As a result of this project, we have created an accurate model that automatically segments liver tumors based on CT scans. In addition, the algorithm provides objective data in the form of a patient’s tumor volume. The automation of the process speeds up the work and the precision of the algorithm supports the human eye.

Image credit: Medical Segmentation Decathlon (medicaldecathlon.com)