Case K: Automated generation of radiology reports from thorax-abdomen CT scans

Background: Radiology is the cornerstone of medical imaging, and is important to diagnose various type of diseases, such as cardiac diseases and cancer. But radiology is going towards a workforce shortage. In the EU, there is an average of 127 radiologist per million of inhabitants, with 45% of these radiologists (in 2022) 51 years or older. This means that in the coming decade, almost 45% of the EU-based radiologists will retire, which will lead to a huge workforce shortage. The number of new cancers on the other hand, will also increase from 3 million in the EU in 2022 to 3.7 million cases in 2050. This means that more cancer patients need to be screened with a decreasing number of radiologists. For CT scans, radiologists need to analyze every slice to detect if there are lesions, tumors or other abnormalities. Especially for oncology patients, the exact quantification of the stage of the cancer is important, since these stages determine the required treatment. But especially for oncology patients, there is a lot of extra work, since the radiologist must compare the lesions in the current scan with the previous scan to determine if there is tumor regression or metastasis (cancer spreads from the primary origin to another location). With the shortage of radiologists and the increase of CT scans, the radiologists must do this analysis in an increasingly short time, which increases the probability of missing crucial abnormalities. It is therefore important that the radiologists can do the analysis of CT scans efficiently.

Objective: The goal of this PhD is to integrate the AI system better in the workflow of the radiologists, by writing a report of the CT scans that the radiologists can use as a basis to do his analysis. The report should clearly state what has been spotted in the CT scan and where (which slice) this has been seen. This speeds up the analysis and decreases the number of false negatives.