TIANJIN -- A team of Chinese scientists have designed an artificial intelligence (AI) tool that predicts hard-to-identify tumor origins with an accuracy rivaling or even surpassing human pathologists.

Cancer of unknown primary site poses challenges to clinicians due to its elusive nature. The CUP, which accounts for 3 percent to 5 percent of all cancers diagnosed in humans, tends to be malignant, with only 20 percent of CUP patients achieving a median survival of 10 months.

The researchers, led by those from Tianjin Medical University and the First Affiliated Hospital of Zhengzhou University, developed a deep-learning method for tumor origin differentiation, trained via cytological images from 57,220 cases at four Chinese hospitals.

The tool called TORCH can identify malignant tumors in fluids accumulated in chest and abdomen and predict their origins, according to the study published this week in the journal Nature Medicine.

TORCH achieved a prediction accuracy for primary tumor origins of 82.6 percent, significantly improving the diagnostic scores compared with four human pathologists, according to the study.

Also, the treatment protocol given in conforming with TORCH-predicted origins resulted in an overall survival of 27 months as against 17 months for those who were administrated discordant treatment.

The study highlighted the AI system's potential as a valuable ancillary tool in clinical practice, although further validation in randomized trials is demanded, said the researchers.