MR CAD boosts diagnostic performance for prostate cancer

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
 - business, upward

Radiologists who used an internally developed MR computer-aided diagnosis (CAD) system improved their performance in the differentiation of benign from malignant lesions at 3T multiparametric MRI, according to a study published online Nov. 30 in Radiology. Performance gains were more pronounced for less experienced reviewers.

As researchers and radiologists explore various advantages and disadvantages of T2-weighted MR imaging and functional MR, multiparametric MR has demonstrated its utility in the detection, localization and characterization of prostate cancer. However, the technique requires a high level of radiologist experience and is prone to observer variability.

Thomas Hambrock, MBChB, from the department of radiology at Radboud University Medical Centre in Nijmegen, the Netherlands, and colleagues designed a study to evaluate the effect of CAD on reader performance in the differentiation of benign from malignant prostate lesions at 3T multiparametric MR imaging.

Six less-experienced radiologists who had interpreted fewer than 50 prostate MR exams and four experienced radiologists who had read more than 100 prostate MR exams reviewed multiparametric 3T MR data for 34 prostate cancer patients. Images were first read without CAD and then reviewed with CAD software. Radiologists received CAD training prior to the study.

The researchers annotated and evaluated 120 benign lesions (64 in the peripheral zone and 56 in the transition zone) and 86 malignant lesions (67 in the peripheral zone and 19 in the transition zone). Histopathologic tumor maps served as ground truth.  

Radiologists were provided with predefined regions of interest (ROI) and estimated the likelihood of malignancy on a scale from 0 to 100 percent for each ROI. Then the ROI CAD likelihood was displayed and the readers entered an additional likelihood of malignancy with CAD.

Without CAD, the less-experienced radiologists had an overall area under the curve (AUC) of 0.81. AUC was 0.86 for the peripheral zone and 0.72 for the transition zone. Experienced readers had an overall AUC of 0.88 without CAD, with AUCs of 0.91 and 0.81, respectively, for the peripheral zone and transition zone.

The addition of CAD resulted in significant improvements in lesion discrimination for less experienced readers, and allowed this group to achieve performance on par with experienced radiologists.

Overall AUC rose to 0.91 for less-experienced readers with the addition of CAD. AUC increased to 0.95 in the peripheral zone and 0.79 in the transition zone. Overall AUC increased to 0.91 for experienced readers. Peripheral zone AUC was 0.93, and transition zone AUC was 0.82.

“Therefore, CAD appears to be a promising method for implementation into the routine clinical environment for the characterization of lesions suspicious for cancer at MR imaging assessment of prostate cancer,” concluded Hambrock et al.