Cirrhosis detection from CT scans

Table of Contents

Hepatic cirrhosis automated analysis?

Hepatic cirrhosis, also known as liver cirrhosis, is an advanced liver disease. It affects about 1 in 400 adults in the US. The standard examination for patient diagnosis and treatment is abdominal computed tomography (CT). CT images’ segmentation is a very useful clinical tool, providing myriad information. However, an accurate assessment of the liver’s condition takes time. It depends on the user (hence lacks reproducibility) and may be inaccurate. In one of our projects, we addressed these research gaps and suggested a complete, repeatable method for cirrhosis detection from CT scans.

Clinically inspired features and radiomics ones

In our approach, we benefit from clinically inspired features that reflect the patient’s characteristics. Radiologists often investigate them during the screening process. Among others we focused on the volume distribution within the CT scan, liver bluntness or the liver surface nodularity.

In the next step we coupled those clinical features with the radiomics ones. These features go beyond what the human eye can easily see and describe the image in more detail. We extracted them from the liver, and from the suggested region of interest which captures the liver’s boundary. Afterwards, we selected a subset of the most discriminative features to enhance the interpretability of the system.

Introducing the liver cirrhosis algorithm

We introduced an algorithm to extract the region of interest corresponding to the rectified boundary of the liver, from which the radiomic features could be extracted. We hypothesized that the features that relate to the various stages of fibrosis are manifested near the liver’s boundary. As a result, we proposed a flexible machine learning system that automatically classifies CT scans (cirrhotic vs. non-cirrhotic). Also, it identifies the most important features for analysis.

the graphic featured a flowchart of the cirrhosis detection from CT scans

Patient cohorts

To quantify the capabilities of the algorithm, we followed rigorous cross-validation over two patient cohorts:

241 portal venous CT scans acquired in a clinical trial,
public benchmark included 32 portal venous CT scans obtained for 32 healthy potential liver donors.

Our study had two goals. Firstly, we wanted to test if our feature extractors can accurately classify portal venous CT scans as cirrhotic or non-cirrhotic. Secondly, we wanted to prove our classification pipeline is adaptable for various uses.

The result: cirrhosis detection from CT scans

Our team reports the detailed results in the below paper.

Krzysztof Kotowski, Damian Kucharski, Bartosz Machura, Szymon Adamski, Benjamín Gutierrez Becker, Agata Krason, Lukasz Zarudzki, Jean Tessier, Jakub Nalepa, Detecting liver cirrhosis in computed tomography scans using clinically-inspired and radiomic features, Computers in Biology and Medicine, Volume 152, 2023, 106378, ISSN 0010-4825, https://doi.org/10.1016/j.compbiomed.2022.106378.