RSNA 2016

Abstract Archives of the RSNA, 2016


RC607-03

The PI-RADS Version 2 Lexicon: Application by Radiologists Inexperienced in Prostate MRI Interpretation

Thursday, Dec. 1 9:20AM - 9:30AM Room: E450B



Awards
Student Travel Stipend Award

Laura M. Leonards, MD, Los Angeles, CA (Presenter) Nothing to Disclose
Nisha Alle, MD, Los Angeles, CA (Abstract Co-Author) Nothing to Disclose
Taylor J. Choy, MD, Los Angeles, CA (Abstract Co-Author) Nothing to Disclose
Daniel J. Margolis, MD, Los Angeles, CA (Abstract Co-Author) Nothing to Disclose
Patrick J. Pan, MD, Los Angeles, CA (Abstract Co-Author) Nothing to Disclose
Andrew B. Rosenkrantz, MD, New York, NY (Abstract Co-Author) Nothing to Disclose
Hyung J. Kim, PhD, Los Angeles, CA (Abstract Co-Author) Nothing to Disclose
PURPOSE

PI-RADS version 2 (v2) includes a detailed lexicon to guide scoring a lesion’s level of suspicion. While the lexicon has been evaluated among experienced radiologists, a key intended benefit of PI-RADS v2 is to assist interpretation by radiologists without prostate MRI expertise. Thus, our aim was to evaluate the performance of radiologists inexperienced in prostate MRI interpretation in applying the PI-RADS v2 lexicon.

METHOD AND MATERIALS

Four radiology residents without prior prostate MRI training evaluated 40 prostate MRI exams. Readers were provided screen captures indicating the location of one specific lesion per case (20 in PZ; 20 in TZ), along with the PI-RADS v2 document. Readers scored the specified lesion for a wide array of lexicon components. These exams had previously been evaluated by six expert prostate MRI radiologists as part of a national multi-center reproducibility study; experts’ consensus readings served as reference. Reader-averaged percent agreement with the reference was computed for the lexicon features (considered excellent when >80%).

RESULTS

In PZ, novice radiologists’ agreement with the expert-derived reference was excellent (84%-90%) for features related to DWI (focal shape; marked high b-value hyperintensity; marked ADC hypointensity; DWI >3), though moderate (75%) for DCE (+). In TZ, agreement was excellent for T2 encapsulation (86%), though moderate (61%-78%) for other T2WI features (circumscribed shape; lenticular shape; heterogeneity; moderate hypointensity; T2 >3) and moderate (74%-81%) for DWI features. Agreement for PI-RADS >3 was 89% in PZ and 75% in TZ. Kappa values were also generally better for PZ than for TZ features (average kappa 0.57 and 0.32 respectively), with moderate to substantial agreement for all PZ features except DCE (fair), but nonsignificant slight agreement for heterogeneity, intensity, and invasiveness, and negative kappa for lenticular, in the PZ. For overall PI-RADS >3, kappa was 0.61 in PZ and 0.40 in TZ.

CONCLUSION

Novice radiologists performed reasonably well using the PI-RADS v2 lexicon, achieving excellent agreement with expert readers in PZ for DWI and overall PI-RADS >3. However, performance was weaker for DCE in PZ and for numerous TZ features.

CLINICAL RELEVANCE/APPLICATION

The results are encouraging regarding novice radiologists’ ability to apply PI-RADS v2 lexicon in practice. Further education should target DCE in the PZ and textural T2-related features in the TZ.