Advertisement
For clinical diagnosis of prostate cancer, radiologists typically use multiparametric MRI, which produces a more detailed picture of the prostate gland than a standard one.
Results are expressed as a PI-RADS score (Prostate Imaging-Reporting and Data System), with a higher score meaning more chances of clinically significant cancer to be present.
However, classifying lesions, or tissue abnormalities, using the PI-RADS score has limitations, according to the study’s authors.
Related Articles
Advertisement
For developing the model, the researchers trained a convolutional neural network (CNN) — a type of artificial intelligence (AI) used primarily for image recognition and processing — to predict clinically significant prostate cancer from multiparametric MRI.
The model’s performance was compared with that of abdominal radiologists in a group of patients who had undergone MRI but without known clinically significant prostate cancer.
“In a retrospective study of 5,215 patients (5,735 examinations) who underwent multiparametric MRI for prostate cancer evaluation, (AI-based) model performance in clinically significant prostate cancer detection was not different from that of experienced radiologists,” the authors wrote.
They said the model could be used as an adjunct to radiologists to improve prostate cancer detection.
“I do not think we can use this model as a standalone diagnostic tool. Instead, the model’s prediction can be used as an adjunct in our decision-making process,” Takahashi said.
A Lancet Commission on prostate cancer projected that between 2020 and 2040, cases around the world could more than double and deaths could increase by 85 per cent. Low- and middle-income countries are expected to bear an overwhelming brunt of the spike, it said.
The commission called for evidence-based interventions, including early detection and diagnosis, to help save lives and prevent ill health from prostate cancer in the coming years.