AI model helps diagnose ovarian cancer

Chinese scientists have developed an artificial intelligence-based model that facilitates the low-cost, easily accessible and accurate diagnosis of ovarian cancer.

A report on the research was published in the international science journal The Lancet Digital Health last week.

Ovarian cancer is the most lethal gynecological malignancy, but timely diagnosis is difficult due to a lack of effective biomarkers, said Liu Jihong, a professor at Sun Yat-sen University's cancer center, who participated in the research with partners from several Chinese universities and hospitals.

Laboratory tests are widely applied in clinical practice, and some have shown diagnostic and prognostic relevance to ovarian cancer.

The researchers aimed to systematically assess the value of routine laboratory tests on the prediction of ovarian cancer, and develop a robust and generalized AI model to assist with diagnosis, she said.

In this study, the team collected laboratory tests and clinical features of 10,992 women with or without ovarian cancer admitted to three hospitals in China from January 2012 to April 2021.

Researchers used a sophisticated system that considers multiple factors to predict the risk of a health condition.

They combined estimations from 20 different AI models that analyzed data from 51 lab test items and the age. When they tested the model at Tongji Hospital of Huazhong University of Science and Technology, it showed high accuracy. Similar good results were seen at Zhejiang University and Liu's center.

This model, according to Liu, outperformed traditional tumor markers in accuracy and sensitivity in detecting ovarian cancer, especially during its early stages.

The new model provides clinicians with a tool to assist with diagnosis, especially in routine health examinations or primary medical facilities with limited clinical experience in gynecological oncology.

In addition, the lab tests included in the model cost less than 1,000 yuan in total ($140), cheaper than a routine female tumor marker panel test plus an imaging examination, said Cai Guangyao, a resident doctor of the Sun Yat-sen University Cancer Center.

Ovarian cancer has a low prevalence in the population, but lacks typical early-stage symptoms, Liu said.

Within the healthcare system, primary care is often the first point of contact for a patient, but clinicians often fail to identify those with early-stage disease due to the absence of clear symptoms.

In China, less than 48 percent of ovarian cancer patients were reported to be diagnosed in the early stages, and the five-year survival rate of ovarian cancer was approximately 40 percent.

With the novel model, doctors will be able just to input the lab test values, and the model will output an estimated probability of having ovarian cancer.