Approximately 5% of cases of metastatic cancer are of unknown primary origin, where cancer cells are typically found dispersed in the peritoneal or pleural fluid, making treatment planning challenging. Diagnosis often requires expert pathologists to compare cancer cell characteristics from detailed microscopic images and perform biopsies from suspected organs of the patient.
A research team from the People’s Republic of China’s Tianjin Medical University and Zhejiang University has developed an artificial intelligence tool called tumor origin differentiation using cytological histology (TORCH). This tool collects images of cancer cells and cells from other diseases from hospitals and cancer research institutions in China, including Tianjin Medical University Cancer Institute, Zhejiang University School of Medicine Hospital, Suzhou University School of Medicine Hospital, and Yantai Yuhuangding Hospital. They gathered and filtered over 50,000 images, dividing them into approximately 30,000 training images and the rest for testing purposes.
Experimental results show that TORCH has an accuracy of about 83% in identifying the primary origin of cancer cells in the test data set. Moreover, when considering the top 3 predictions each time, there is a high probability of up to 98.9% that the correct answer will be among the top 3 predicted results.
The research team hopes that these satisfactory results demonstrate the potential of using artificial intelligence to aid in diagnosing the origin of cancer in challenging cases. However, these findings are based only on patient data in China. Testing TORCH with external databases, such as The Cancer Genome Atlas (TCGA), showed a slightly lower accuracy ranging from 70-88%.
This research was published in Nature Medicine. DOI:10.1038/s41591-024-02915-w.
TLDR: A research team in China developed an AI tool, TORCH, that can accurately identify the primary origin of cancer cells in challenging cases, showing promise for using AI in cancer diagnosis.
Leave a Comment