A computer-aided diagnosis (CAD) system may improve the characterization of prostate lesions at multiparametric MRI by increasing reading specificity, according to a study published online Mar. 5 by Radiology.
Though multiparametic MRI has demonstrated promise for the detection of prostate cancer, characterization of focal prostate lesions is still difficult, particularly for less experienced readers. More objective criteria for malignancy have been developed to try to solve the problem, yet uncertainty remains about the solution’s effectiveness. An alternative fix, however, could be the utilization of CAD systems.
Lead author Emilie Niaf, PhD, of the Université de Lyon in France, and colleagues designed a CAD system to assess the likelihood of visible lesion malignancy at multiparametric MRI that combines T2-weighted, diffusion-weighted and dynamic contrast-enhanced imaging. The researchers’ study investigated their CAD system’s proficiency in improving focal prostate lesion characterization at muliparametric MRI by readers with varied experience levels.
The study’s authors reviewed 30 consecutive 1.5-T multiparametric MRI studies performed before radical prostatectomy in patients between September 2008 and February 2010. Assessment was separated into three sessions, during which twelve readers used a five-level subjective score to investigate malignancy likelihood in 88 predefined peripheral zone lesions.
Less experienced readers showcased significantly lower mean area under the receiver operating characteristic curve (AUC). Seven of the readers bettered their performance between reading sessions one and two, and 12 readers underwent performance improvement between sessions two and three.
The mean AUC for reading session one was 83 percent and 84.1 percent for session two, revealing no significant difference. While the mean AUC for session three was higher than the second at 87.2 percent, there was still no significant difference found.
A subjective score positivity threshold of three demonstrated a specificity of 79 percent in reading session two, which was not significantly different than the 78.7 percent specificity in session one. However, reading session two’s specificity was significantly lower than that of the third session, which was 86.2 percent. Lastly, the sensitivity of reading session two, 68.4 percent, was significantly higher than session one, but not significantly different from session three.
Despite the moderate improvement of reader performance across the board with CAD assistance, statistical significance was not reached. However, Niaf and colleagues still believe their results are encouraging and that future studies are necessary to establish how CAD systems can be used in daily routine.
“Our results suggest that the main benefit of our CAD system was in improving the correct classification of benign lesions,” they wrote. “This is an interesting perspective because up to 75% of focal lesions seen at multiparametric MR imaging are benign and the false-positive rate at multiparametric MR imaging can be as high as 40%, even for experienced readers. Recognizing a portion of these false-positives as formally benign could avoid unnecessary biopsy when multiparametric MR imaging is used as a triage test or improve patient selection for focal therapy.”