Based on findings from a multicenter study of over 1,600 patients, researchers at the European Congress of Radiology suggest the inclusion of autonomous artificial intelligence (AI) triage could facilitate up to a 75 percent reduction in prostate MRI reading workload.
The use of an autonomous artificial intelligence (AI) model for clinically significant prostate cancer (csPCa) detection on magnetic resonance imaging (MRI) demonstrated better accuracy than radiologist readers at three different centers, a finding that could significantly alleviate increasing prostate MRI worklist volume, according to research presented recently at the European Congress of Radiology.
For the study, which involved a total of 1,612 patients from three different facilities, researchers utilized AI as a first reader of prostate MRI scans with a subsequent certainty score determining the next step in the detection pathway. The study authors noted that findings on MRI with high certainty for clinically significant prostate cancer (csPCa) did not proceed to the radiologist worklist whereas cases of “uncertain” findings on MRI were triaged for radiologist review.
The researchers employed a non-inferiority specificity margin of -0.05 to calculate accuracy with the AI assistance pathway whereas radiologist accuracy was determined by comparing their PI-RADS 4 and higher assessments with biopsy findings.
At the first center, involving 689 patients, researchers noted 70 percent accuracy with AI assistance vs. 65 percent with radiologist-only reading. In a cohort of 723 patients at the second facility, the study authors noted a 13 percent higher accuracy with AI assistance (75 percent vs. 62 percent), and nearly equivalent accuracy (85 percent vs. 83 percent) at a third center with 200 patients.
“The AI-supported PCa detection pathway provided a more efficient diagnosis compared to the radiologist-only pathway while maintaining a similar accuracy and non-inferior specificity,” wrote lead study author Stefan J. Fransen, MSc, who is affiliated with the Department of Radiology at University Medical Center Groningen in Groningen, Netherlands, and colleagues.
Given workforce shortages in the field and increasing imaging volume, the researchers said the utility of AI could have a significant impact in reserving the more complex diagnostic cases for radiologist review.
“This study shows the feasibility of using uncertainty to triage PCa suspicious patients to AI reading or (radiologist) assessment,” added Fransen and colleagues. “Based on our findings, AI-supported uncertainty-based diagnosis could reduce the radiologist workload by 50% to 75% without losing diagnostic accuracy.”
The researchers cautioned that subsequent prospective studies are necessary to provide more clarity on AI performance and the impact upon radiologist workloads.
Reference
1. Fransen S, Bosma J, van Lohuizen Q, et al. A multi-center validation of prediction uncertainty to select patients for prostate cancer AI diagnosis. Presented at the European Congress of Radiology, February 28-March 3, 2024, Vienna, Austria. https://www.myesr.org/congress/
MRI-Based AI Radiomics Model Offers 'Robust' Prediction of Perineural Invasion in Prostate Cancer
July 26th 2024A model that combines MRI-based deep learning radiomics and clinical factors demonstrated an 84.8 percent ROC AUC and a 92.6 percent precision-recall AUC for predicting perineural invasion in prostate cancer cases.
Systematic Review: PET/MRI May be More Advantageous than PET/CT in Cancer Imaging
July 18th 2024While PET/MRI and PET/CT had comparable sensitivity for patient-level regional nodal metastases and lesion-level recurrence, the authors of a systematic review noted that PET/MRI had significantly higher accuracy in breast cancer and colorectal cancer staging.