An emerging deep learning model for prostate magnetic resonance imaging (MRI) may provide enhanced clarity for PI-RADS 3 assessments.
For a new retrospective multicenter study, recently published in Insights into Imaging, researchers evaluated deep learning models with MRI-based double channel attention modules (AttenNet) in patients with initial PI-RADS 3 evaluations. The deep learning models were initially trained with 1,144 PI-RADS 1-2 and 4-5 cases, and subsequently retrained with PI-RADS 3 cases drawn from three facilities, according to the study. The study authors noted that external validation testing of the deep learning models was performed on 185 PI-RADS 3 cases from three different institutions.
The researchers found that the external validation testing of the deep learning models had an average 89.3 percent area under the receiver operating characteristic curve (AUC) in detecting prostate cancer (PCa). The study authors also noted an average 87.65 AUC for diagnosing clinically significant prostate cancer (csPCa) in PI-RADS 3 cases from external validation testing.
“These findings suggested that the proposed AttenNet models may be a promising tool to aid the precise risk stratification of PI-RADS 3 patients. … In contrast to some previous radiomics studies based on the manual segmentation and annotation of prostatic lesions for PI-RADS 3 patients, the (deep learning model in the) present study can automatically mine the deep features of the lesions and their periphery and therefore is more conveniently applied to clinical practice,” wrote lead study author Jie Bao, M.D., who is affiliated with the Department of Radiology at the First Affiliated Hospital of Soochow University in Suzhou, China, and colleagues.
The study authors also noted that the deep learning models downgraded 62.2 percent of PI-RADS 3 lesions at one facility and 78.1 percent at another institution involved in external validation testing.
Three Key Takeaways
1. Improved risk stratification for PI-RADS 3 cases. The AttenNet deep learning model demonstrated an 89.3 percennt AUC for detecting prostate cancer (PCa) and an 87.65 percent AUC for clinically significant prostate cancer (csPCa) in PI-RADS 3 cases, suggesting it could enhance risk stratification and improve diagnostic precision.
2. Potential to reduce unnecessary biopsies. The model downgraded 62.2 to 78.1 percent of PI-RADS 3 lesions in external validation, indicating that it may help reduce unnecessary biopsies by improving specificity and identifying cases where immediate biopsy may not be needed.
3. Automation and clinical integration. Unlike traditional radiomics methods that rely on manual lesion segmentation, the study authors noted the deep learning model can automatically extract lesion features and the lesion periphery, making it more practical and convenient for routine clinical application in prostate MRI interpretation.
“ … The AttenNet model has great potential to improve the (specificity) of the diagnosis of csPCa based on MRI images and therefore, to decrease the risk-benefit ratio of biopsy for PI-RADS 3 patients. The AttenNet model can be regarded as a triage test to decide which patients should undergo biopsy and which patients could safely avoid immediate painful biopsy,” emphasized Bao and colleagues.
For the detection of PCa and csPCa, the deep learning models also had a satisfactory AUC for the D-max parameter, which aids in differentiating PI-RADS 4 and PI-RADS 5b cases and offered reliable discrimination for external validation cases involving high, moderate, and low prostate-specific antigen (PSA) levels, according to the researchers.
(Editor’s note: For related content, see “Can Deep Learning Radiomics with bpMRI Bolster Accuracy for Prostate Cancer Prognosis?,” “Can Generative AI Facilitate Simulated Contrast Enhancement for Prostate MRI?” and “Study Affirms Low Risk for csPCa with Biopsy Omission After Negative Prostate MRI.”)
In regard to study limitations, the authors acknowledged variable sample sizes at the participating centers and the retrospective nature of the study, emphasizing the need for prospective multicenter research to provide further validation of the study findings.
