A recent study published in the journal Radiology reveals that an artificial intelligence (AI) model can match the performance of experienced radiologists in detecting clinically significant prostate cancer using magnetic resonance imaging (MRI). Prostate cancer, the second most common cancer among men, is typically diagnosed with multiparametric MRI. However, interpreting these images can be challenging, with the accuracy often depending on the radiologist’s experience.
Traditional AI models require radiologists or pathologists to annotate each lesion, which is a time-consuming and resource-intensive process. This limitation can hinder the development and scalability of AI models. The study, led by Dr. Jason C. Cai from the Mayo Clinic, sought to overcome this challenge by developing a deep learning model that predicts the presence of clinically significant prostate cancer (csPCa) using patient-level labels, without needing information about the exact tumor location.
The research team reviewed data from patients who underwent MRI at the Mayo Clinic between 2017 and 2019. Out of 5,735 examinations, 1,514 showed csPCa. The AI model was then tested on a set of 400 exams, along with an external dataset of 200 exams. Remarkably, the AI model’s performance matched that of experienced abdominal radiologists in detecting csPCa.
The study also found that combining the deep learning model’s predictions with the radiologists’ interpretations resulted in even better performance than using AI alone. This suggests that the AI model could serve as an effective assistant to radiologists, helping them to improve detection rates and reduce false positives.
However, the researchers caution against using the AI model as a standalone diagnostic tool. Co-author Dr. Naoki Takahashi emphasized that the model’s predictions should be used as an adjunct in the decision-making process, rather than replacing human judgment.
Moving forward, Mayo Clinic researchers are expanding their dataset and plan to conduct prospective studies to observe how radiologists interact with the AI model in real-world scenarios. They aim to assess whether the combination of AI and radiologist input leads to better diagnostic accuracy compared to radiologists working alone.
Additionally, the field is shifting towards biparametric prostate MRI (bpMRI), which reduces scan duration, costs, and risks associated with contrast agents. The AI model’s adaptation to bpMRI could significantly enhance its utility and impact, making MRI screening for csPCa more accessible and cost-effective.
This research represents a promising step toward improving the early and accurate diagnosis of prostate cancer, ultimately aiming to enhance patient outcomes through the integration of advanced AI tools in clinical practice.
