Commercially available AI significantly improves prostate MRI report consistency
An artificial intelligence-based software developed to improve prostate cancer diagnostics recently proved its utility for reducing clinical workloads and improving patient outcomes.
The software—mdprostate (mediaire GmbH, Berlin, Germany)—is commercially available and CE marked. Experts recently integrated the tool into their organization’s PACS to apply it to a group of patients who underwent multiparametric prostate MRI to compare its performance to that of the radiologists who completed the exams’ initial interpretations. The team found that the software’s performance was in line with the rads, indicating that it could be beneficial in addressing issues with variability in reporting, experts involved in the analysis suggested.
“By providing objective assessments and standardizing lesion detection and classification, AI has the potential to augment radiologists’ performance throughout the PCa diagnostic pathway,” corresponding author Nadine Bayerl, with the Institute of Radiology at University Hospital Erlangen in Germany, and co-authors noted. “Recent advances in deep learning algorithms, facilitated by larger labeled datasets, improved computing hardware, and refined training techniques, have led to several studies highlighting the diagnostic value of deep learning algorithms in prostate imaging.”
For the study, the team retrospectively applied the tool to 123 prostate MRI exams. The software was tasked with automatically segmenting the prostate, calculating prostate volume and classifying lesions according to PI-RADS scores.
For exams with PI-RADS scores greater than 2, the software yielded 100% sensitivity, ruling out all cancers—both clinically significant and insignificant—for all lesions that fell below the threshold. For scores of 4 or higher, the tool achieved a sensitivity of 85.5 % and specificity of 63.2 % for clinically significant cancers. It yielded an AUC of 0.803 for detecting cancers of any grade.
“In practical terms, these results indicate that when a case falls below the PI-RADS ≥ 2 cutoff, clinicians can rule out malignancy with a high degree of confidence,” the authors explained. “This capability is particularly valuable in clinical decision-making, as it allows for the safe avoidance of unnecessary biopsies or further invasive procedures in these patients.”
The team suggested that the fully automated tool could have real utility for improving clinical workflows and improving patient care, though their work does need to be validated in prospective studies to fully take advantage of the benefits.
The study abstract is available here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She joined Innovate Healthcare in 2021 and has since put her unique expertise to use in her editorial role with Health Imaging.