AI technology detects 15 liver cancer cases missed in clinical trials

A liver cancer diagnostic AI model developed by a team of Chinese scientists found 15 cancer cases that were initially missed during a two-month real-world prospective clinical trial, helping patients receive timely surgical treatment or medication, as reported by China Science Daily.

Early detection of liver cancer is challenging. Because the lesions are small and easily obscured by cirrhosis, fatty liver, and the complex internal anatomy of the liver, even experienced doctors can miss lesions on contrast-enhanced CT scans.

This is even more difficult when cancer from other parts of the body metastasises to the liver. Doctors’ attention may be focused on the primary lesion, causing very small liver metastases to go unnoticed.

Alibaba’s DAMO Academy, in collaboration with Shengjing Hospital of China Medical University and several other institutions, developed an AI model for liver cancer diagnosis called DAMO LiON, which is capable of identifying very small liver cancer lesions from CT images. The model can not only accurately identify primary liver cancer, but also detect liver metastases that are often missed.

Experimental results showed that DAMO LiON’s accuracy in identifying malignant tumours is higher than that of general radiologists. When doctors used the AI model to assist in reading scan results, reading time was reduced by 27 per cent, while sensitivity to malignant tumours increased by 11.5 per cent, effectively reducing missed diagnoses. With AI assistance, junior doctors were able to work at a level equivalent to senior doctors.

In the two-month real-world prospective clinical trial, DAMO LiON read contrast-enhanced CT scan results from more than 10,000 patients, helping doctors detect 15 initially missed liver metastasis cases and prompting changes to those patients’ treatment plans.

Liver metastases detected by DAMO LiON generally had characteristics of being small, faint, and not centrally located. With an average diameter of around 1 centimetre, the lesions had low contrast against liver tissue, or were located in relatively uncommon anatomical positions.

The research was recently published in the journal Nature Medicine.

DAMO Academy has also developed AI screening models, including DAMO PANDA for pancreatic cancer, DAMO GRAPE for gastric cancer, and DAMO COCA for colorectal cancer.