Automated diagnosis of prostate cancer location and form by artificial intelligence in multiparametric MRI
INTRODUCTION
MR-targeted biopsy improves accuracy of transrectal prostate biopsy. With increasing use of 3T-MRI, prostate MRI scans for pre-biopsied patients become more important. On the other hand, MRI diagnosis requires multi-sequential interpretations of multi-slice images by fewer radiologists, while number of prostate cancer patients is rapidly increasing. To reduce such burden, artificial intelligence(AI)-based diagnosis is expected to be a critical technology. Accurate detection of cancer location enhances accuracy of prostate biopsy. However, few studies have focused on diagnosis of cancer location and form. We present AI-based diagnostic method of prostate cancer location and form in multiparametric MRI.
METHODS
The study enrolled 15 patients who underwent radical prostatectomy between April 2008 and August 2017 in our institute. We labeled cancer area on peripheral zone (PZ) on MRI images comparing MRI and pathological mapping on radical prostatectomy specimens. Likelihood maps were drawn and segmented into physiologically shaped regions by superpixel method. Likelihood maps consist of pixels which take cancer likelihood value computed from T2 weighted (T2WI), apparent diffusion coefficient (ADC) and diffusion weighted MRI-based texture features. Cancer location is determined by identifying cancer regions. We evaluated diagnostic performance by area weighted sensitivity and specificity.
RESULTS
Sensitivity and specificity for our approach were 0.89 and 0.87 respectively (p
CONCLUSION
Non-linear segmentation of likelihood maps can be successfully applied to AI-based automatic diagnosis of prostate cancer location and form.
Funding: none