Improving Prostate Cancer Screening With an AI Model
Madhur Nayan, MD, PhD, discusses a new machine learning model developed to enhance prostate cancer screening.
Madhur Nayan, MD, PhD, assistant professor of urology and population health at NYU Langone Health, discusses a new machine learning model, ProMT-ML, developed to enhance prostate cancer screening by predicting the likelihood of an abnormal prostate MRI using electronic health record data.
Prostate MRI is a critical tool that aids clinicians in both reducing the underdiagnosis and overdiagnosis of prostate cancer. However, its widespread use is currently limited by factors such as resource constraints, high costs, and the requirement for specialized expertise. This can lead to prolonged waiting times for some patients and potentially unnecessary biopsies for others who haven't undergone an MRI.
To address these limitations, experts aimed to create a model that could help prioritize patients who are more likely to benefit from a prostate MRI. The ProMT-ML model analyzes various factors readily available in electronic health records, including age, prostate specific antigen (PSA) levels, prostate size (volume), and Body Mass Index (BMI). By integrating each of these clinical parameters, the model is able to offer a more nuanced prediction of an abnormal MRI compared with the current standard of relying solely on PSA levels.
Nayan explains that the existing approach often leads to unnecessary MRIs in patients with elevated PSA but ultimately normal MRI findings. ProMT-ML seeks to improve this by combining different clinical parameters to better identify individuals with a higher probability of an abnormal prostate MRI.
“What we have tried to do is combine different clinical parameters to better predict which patients have an abnormal prostate MRI. We have taken factors like age into account, their body mass index, and this provides more information than just the PSA alone in predicting that abnormal prostate MRI,” Nayan says.
The study demonstrated that the most accurate iteration of their model, ProMT-ML, achieved an accuracy of 75% in predicting abnormal MRI results, suggesting that the model has the potential to optimize the utilization of prostate MRI resources, potentially reducing wait times for those who truly need the imaging and minimizing unnecessary biopsies in individuals less likely to have significant findings. While the model focuses on predicting MRI outcomes, Nayan notes that a predicted normal MRI often correlates with clinically insignificant disease, suggesting its broader utility in guiding downstream diagnostic pathways.
