br MRI Interpretation br Readers were

MRI Interpretation
Readers were instructed to use RadiAnt DICOM Viewer to view images (14). All 135 sequences were randomized per reader and provided in this order with a read-out form designed with Microsoft Access (15). Within this programmed form, shown in Figure 2, readers recorded up to four detected lesions per sequence. Readers identified suspicious regions on each sequence utilizing visual patterns characteristic for tumor as outlined in PI-RADSv2 (5). For each lesion detected, read-ers recorded lesion location (zone and side of prostate), anno-tated the provided MRI sequence to delineate the diameter of the identified lesion, and uploaded a screenshot. All data were recorded in a linked Microsoft Access database.
Pathology Correlation
After mpMRI, case patients underwent robotic-assisted radi-cal prostatectomy. For optimal image-pathology correlation, all prostate specimens were processed with patient-specific MRI-based three-dimensional printed molds with sections
TABLE 1. Multiparametric MRI Acquisition Parameters Given for Prostate Imaging at 3-Tesla (3T) with Use of an Endorectal Coil.
Multiparametric MRI Sequence Parameters at 3T
Sequences acquired include T2-weighted, diffusion-weighted imaging (DWI) that includes apparent diffusion coefficient map calculation, and a high b-value sequence, and dynamic contrast enhanced (DCE) magnetic resonance imaging (MRI). * For ADC map calculation. Five evenly spaced b-values (0 750 s/mm2) were used. y b = 2000 s/mm2. z DCE MRI obtained before, during, and after a single dose of gadopentetate dimeglumine 0.1 mmol/kg at 3 mL/s. Each sequence was obtained at 5.6 s intervals.
cut spanning apex to AZD8931 (16). One highly experienced geni-tourinary pathologist annotated the location of cancerous regions and lesion-specific Gleason scores ( 3 + 3) were assigned. A prostate mpMRI-focused research fellow per-formed radiologic-pathologic correlation using screenshot annotations of detected lesions from each reader. Magnetic resonance lesions detected were correlated to the pathology maps using visible prostate landmarks and lesion morphology. Lesions outlined on pathology but not detected by any read-ers were noted as false negative reads in subsequent analysis.
Statistical Analysis
Lesions were considered detected by a reader if a screenshot was uploaded showing clear annotation of the lesion on the sequence, with a category assigned. Reader-based lesion detec-tion sensitivity was evaluated for each sequence. Concordance of lesion detection across readers was measured by index of spe-cific agreement (ISA), defined as the conditional probability of a randomly selected individual reader, detecting a lesion in the same location as a blinded, independent reader. Pairwise ISAs were calculated over three pairs of readers and average ISA was reported. Cancer detection rate (CDR) was defined as the pro-portion of true positive lesions among all detected lesions (sum of true positives and false positives) and was calculated as the weighted average of reader CDRs. For example, a single unique lesion detected by all readers would be considered as three total lesions. This weighted average serves to minimize the variability of CDRs in categories where the number of lesions was small. All detected lesions were classified based on independent posi-tivity read from all three sequences, lending to seven possible combinations of sequence detection positivity for each reader: T2W+/DWI¡/DCE¡ (T2W-only), T2W¡/DWI+/DCE¡ (DWI-only), T2W¡/DWI¡/DCE+ (DCE-only), T2W +/DWI+/DCE¡ (T2W+ DWI+), T2W+/DWI¡/DCE+ (T2W+ DCE+), T2W¡/DWI+/DCE+ (DWI+ DCE+), and T2W+/DWI+/DCE+ (all positive). Sensitivity and CDR are Writhing number reported for whole prostate (WP), TZ, and PZ, separately.