Multiparametric magnetic resonance imaging of the prostate with computer-aided detection: experienced observer performance study

Objectives

To compare the performance of experienced readers in detecting prostate cancer (PCa) using likelihood maps generated by a CAD system with that of unassisted interpretation of multiparametric magnetic resonance imaging (mp-MRI).

Methods

Three experienced radiologists reviewed mp-MRI prostate cases twice. First, readers observed CAD marks on a likelihood map and classified as positive those suspicious for cancer. After 6 weeks, radiologists interpreted mp-MRI examinations unassisted, using their favourite protocol. Sensitivity, specificity, reading time and interobserver variability were compared for the two reading paradigms.

Results

The dataset comprised 89 subjects of whom 35 with at least one significant PCa. Sensitivity was 80.9% (95% CI 72.1–88.0%) and 87.6% (95% CI 79.8–93.2; p = 0.105) for unassisted and CAD paradigm respectively. Sensitivity was higher with CAD for lesions with GS > 6 (91.3% vs 81.2%; p = 0.046) or diameter ≥10 mm (95.0% vs 80.0%; p = 0.006). Specificity was not affected by CAD. The average reading time with CAD was significantly lower (220 s vs 60 s; p < 0.001).

Conclusions

Experienced readers using likelihood maps generated by a CAD scheme can detect more patients with ≥10 mm PCa lesions than unassisted MRI interpretation; overall reporting time is shorter. To gain more insight into CAD–human interaction, different reading paradigms should be investigated.

Key points

• With CAD, sensitivity increases in patients with prostate tumours ≥10 mm and/or GS > 6.

• CAD significantly reduces reporting time of multiparametric MRI.

• When using CAD, a marginal increase of inter-reader agreement was observed.