Artificial intelligence (AI) and machine learning (ML) can effectively detect and diagnose polycystic ovary syndrome (PCOS), the most common hormone disorder in women that typically occurs between the ages of 15 and 45, according to a new study from the National Institutes of Health (NIH). Researchers systematically reviewed published scientific studies using AI/ML to analyze data to diagnose and classify PCOS and found that AI/ML-based programs were able to successfully detect PCOS.
"Given the substantial burden of underdiagnosis and misdiagnosis of PCOS in the community and its potentially severe consequences, we wanted to determine the utility of AI/ML in identifying patients who may be at risk for PCOS," said study co-author Janet Hall, MD, senior investigator and endocrinologist at the National Institute of Environmental Health Sciences (NIEHS), part of the National Institutes of Health (NIH). "The effectiveness of artificial intelligence and machine learning in detecting PCOS is more impressive than we thought."
Polycystic ovary syndrome occurs when the ovaries fail to work properly, in many cases along with elevated testosterone levels. The disease can cause irregular menstruation, acne, facial hair, or hair loss on the head. Women with PCOS are generally at increased risk for type 2 diabetes, sleep, psychological, cardiovascular and other reproductive disorders such as uterine cancer and infertility.
"Diagnosing PCOS can be challenging given its overlap with other conditions," said the study's senior author Skand Shekhar, MD, an assistant research physician and endocrinologist at the National Institutes of Health. "These data reflect the untapped potential to incorporate AI/ML into electronic health records and other clinical settings to improve diagnosis and care of women with PCOS."
The study's authors recommend combining large population-based studies with electronic health datasets and analyzing common laboratory tests to identify sensitive diagnostic biomarkers that could help diagnose PCOS.
PCOS is diagnosed based on standardized criteria that have evolved over the years and are widely accepted, but generally include clinical features (such as acne, excessive hair growth, and irregular menstruation) as well as laboratory (such as high blood testosterone) and radiologic findings (such as multiple small cysts and increased ovarian size on ovarian ultrasound). However, PCOS is often overlooked because some features of PCOS may coexist with other conditions such as obesity, diabetes, and cardiometabolic disorders.
Artificial intelligence refers to the use of computer-based systems or tools to mimic human intelligence and help make decisions or predictions. ML is a branch of artificial intelligence that focuses on learning from previous events and applying this knowledge to future decisions. Artificial intelligence can process large amounts of different data, such as that obtained from electronic health records, and is therefore an ideal aid in diagnosing hard-to-diagnose conditions such as polycystic ovary syndrome.
The researchers conducted a systematic review of all peer-reviewed studies using AI/ML to detect PCOS published in the past 25 years (1997-2022). With the help of an experienced NIH librarian, researchers identified potentially eligible studies. They screened a total of 135 studies and included 31 in this article. All studies were observational and evaluated the use of AI/ML technologies in patient diagnosis. About half of the studies included ultrasound images. The average age of study participants was 29 years old.
In 10 studies using standardized diagnostic criteria to diagnose PCOS, the detection accuracy ranged from 80% to 90%.
"Across various diagnostic and classification modalities, AI/ML performed extremely well in detecting PCOS, which is the most important conclusion from our study," Shekhar said.
The authors note that AI/ML-based projects have the potential to greatly improve our ability to detect women with PCOS early, thereby saving related costs and reducing the burden PCOS imposes on patients and the healthcare system. Follow-up research with strong validation and testing practices will enable smooth integration of AI/ML with chronic health conditions.