Artificial intelligence (AI), which uses machines or computers to perform tasks typically done by humans, is a hot topic these days. But what can AI really do?
In the case of cancer care, a team from Singapore General Hospital (SGH) and National Cancer Centre Singapore (NCCS) has been studying the use of AI in a research project to see if it can improve how liver cancer is diagnosed.
The difficulty with diagnosing liver cancer
Liver cancer diagnosis is based on CT and MRI scans, as well as clinical findings, and treatment decisions are made before surgery or biopsy is done.
Liver cancer is common in Singapore. It is the third leading cause of cancer death in men and the fifth in women. The best way to improve outcomes would be to diagnose the disease earlier, however, diagnosing liver cancer can be challenging.
Liver cancer diagnosis is based on CT and MRI scans as well as clinical findings, and treatment decisions are made before surgery or biopsy is done. Imaging generates many images of the tumour, which radiologists need to interpret to diagnose and stage the cancer. Apart from liver cancers, liver diseases such as hepatic steatosis are diagnosed by interpreting histological images analysed by pathologists using microscopy.
Using AI to enhance diagnostic and prognostic accuracy
With the aim of addressing these challenges and improving diagnostic and prognostic accuracy, Project RAPIER (Radiology Pathology Information Exchange Resource), a collaboration between A*STAR's Institute of High-Performance Computing (IHPC), SGH and NCCS, was initiated in 2022.
The project pools over 5,000 imaging studies from SGH and NCCS radiology, and pathology reports from SGH for patients with primary liver cancer (hepatocellular carcinoma, HCC), bile duct cancer (cholangiocarcinoma) and some non-cancerous tumours (such as haemangioma) into a data repository, which was annotated and utilised to train AI algorithms to detect and classify liver disease.
RAPIER provides a clinical decision support system for radiologists and pathologists as they interpret radiological and histological imaging, aiding in the diagnoses of conditions like liver cancers and liver steatosis. This rich resource ultimately acts as a real-time second opinion for radiologists and pathologists, improving accuracy and potentially speeding up turnaround time.
"In the age of radiological scanners with high resolution and explosive growth in demand for medical imaging, such a system has the potential to help radiologists diagnose conditions with greater speed and accuracy," explained Clinical Assistant Professor Gideon Ooi, Consultant, Division of Oncologic Imaging, NCCS, and Site-PI for Project RAPIER at NCCS.
"With RAPIER's AI quantification, we can also help pathologists assess biopsy specimens for hepatic steatosis with more precision, reducing variability in individual analysis, allowing for improvements in patient care."
The application of RAPIER
(From left) Clin Asst Prof Gideon Ooi, Consultant, Division of Oncologic Imaging, NCCS, and Clin Asst Prof Tiffany Hennedige, Head, Division of Oncologic Imaging, NCCS, discussing a liver cancer patient's scan.
Project RAPIER has several key objectives, including developing AI tools to accurately label liver lesion images from radiological data, as well as recognising and describing liver lesions. The AI can distinguish between benign and malignant lesions, which could improve the early detection of liver cancer and other diseases affecting the liver. All of this will not only speed up the diagnostic process but also reduce the potential for errors.
Finally, automating analysis of liver imaging over multiple time points paves the way for future research and development of AI tools to better predict progression of liver cancers. This will enable better prognostication and clinical management.
"So far, RAPIER looks promising. It is a tool that we welcome in diagnostic radiology, as it can potentially reduce the resources needed to analyse the scans, while improving the accuracy of diagnoses," said Clin Asst Prof Ooi.
RAPIER is currently being used in the research setting, with plans to implement it in clinical workflows in future phases of the project.
Collaborating to meet clinical needs
Although AI will not replace radiologists or oncologists, it is becoming increasingly clear that AI's role as an enabling support to their work will be indispensable. According to Clin Asst Prof Ooi, it will help streamline processes, enhance diagnosis, and reduce errors.
"AI can certainly support radiologists, but it cannot replace them," said Clin Asst Prof Ooi. "For AI to be truly effective, clinicians must be involved in its training, ensuring that the tools developed are relevant to real-world clinical settings and patients' needs."
Subscribe to our mailing list to get the updates to your email inbox...
