Multiparametric magnetic resonance imaging of the prostate underestimates tumour volume of small visible lesions

Objective To assess the relationship between the volume of the index lesion (IL) measured at multiparametric magnetic resonance imaging (mpMRI; MRIvol) and at radical prostatectomy (RPvol), stratifying it according to Prostate Imaging‐Reporting and Data System (PI‐RADS) score. Patients and Methods We identified 332 men with a positive mpMRI (single lesion with PI‐RADS ≥3) who underwent systematic plus targeted biopsy and subsequent RP at two tertiary referral centres between 2013 and 2018. All mpMRIs were reviewed by experienced radiologists using PI‐RADS scores. The study outcome was to assess the relationship between MRIvol (based on planimetry from MRI sequence best showing tumour) and RPvol (based on tumour involved area of each RP pathology slice). To achieve this endpoint, we performed a multivariable linear regression analysis (LRA) to predict RPvol using PI‐RADS, prostate‐specific antigen level, prostate volume, age, digital rectal examination, Gleason score at MRI‐targeted biopsy, biopsy history and time from mpMRI to RP as covariates. Non‐parametric locally estimated scatterplot smoothing (LOESS) function was used to graphically explore the relationship between MRIvol and RPvol, stratifying for PI‐RADS score. Results Overall, 24%, 49% and 27% of men had visible PI‐RADS 3, 4 and 5 lesions at mpMRI. The median (interquartile range [IQR]) MRIvol and RPvol were 0.67 (0.29–1.76) mL and 1.39 (0.58–4.23) mL. At LRA, MRIvol was significantly correlated with a RPvol underestimation (slope: 2.4, 95% confidence interval [CI] 0.1–46.3). The non‐parametric LOESS analysis showed a non‐linear relationship between MRIvol and RPvol. Significant underestimation was reported across all volumes with the highest differences between MRIvol and RPvol in the low volume range (