Keywords:
Genital / Reproductive system male, Oncology, Pelvis, MR, MR-Diffusion/Perfusion, Image manipulation / Reconstruction, Computer Applications-Detection, diagnosis, Diagnostic procedure, Staging, Cancer, Neoplasia, Pathology
Authors:
S. Monti1, M. Aiello1, M. Inglese1, C. T. Diomaiuti1, A. Ragozzino2, C. Cavaliere1; 1Naples/IT, 2Napoli/IT
DOI:
10.1594/ecr2018/C-1838
Methods and materials
40 foci of PCa,
48 BPH nodules,
and 36 benign PZ from 40 patients who underwent multiparametric MRI of prostate,
to eventually address a target-biopsy,
were evaluated.
MRI exam was performed without endorectal coil with a 3T mMR Biograph scanner.
DWI was performed using 7 b-values (0-2500 s/mm2): classical apparent diffusion coefficient (ADC) map was generated using b values up to 1500 s/mm2; the entire range of b values was used to compute non-Gaussian diffusion coefficient (D) and deviations from normal distribution (K) maps.
Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) parameters (Ktrans,
ve,
kep [TM & SSM] and intracellular water molecular lifetime τi [SSM]) were also determined.
2D regions of interest were manually outlined on the axial T2w that was also used as reference image to coregister other maps.
13 first-order statistical features were extracted for each map.
Feature selection was performed to identify 5 features for each model.
A logistic regression classifiers were used to identify features discriminating clinically significant tumors (PIRADS 4-5) from benign PZ with an intra- and inter-model approach.
The area under the receiving-operating characteristic curve (AUC),
the sensitivity,
the specificity and the accuracy were used to evaluate the performances of models.
The intermodel approach was also tested to discriminate between BPH and other groups.