Matlab 2018b

ML analyses, survival curve visualization, Cox proportional hazards models, and k-fold cross-validation paired sample t-test were conducted in MATLAB 2018b (Mathworks, Natick, MA). mRMR feature selection was conducted using the Feature Selection Library MATLAB toolbox²⁶ . Construction of the online calculator (Supplemental Methods) and all remaining statistics and were computed in RStudio 1.1.456 (RStudio, Boston, MA) with R 3.5.1 (R Foundation, Vienna, Austria) with packages “pROC^{27 (link)},” “compareC^{25 (link)},” “e1071,” and “tableone.”

We used two prospectively-maintained databases: one that linked fusion biopsy to pathology and MRI results, and another that described lesion features on WMP. Custom Matlab 2018b (MathWorks, Natick, MA) scripts were written to extract and compute all data features, also known as covariates when using standard multivariable analysis terminology. All imaging and pathology reports in patient charts were checked manually by four separate authors (SRZ, DCJ, JJY, JB) to ensure accuracy of the extracted data.
In addition to obtaining standard clinicopathologic features used in routine diagnosis and assessment—age, PSA-related features, prostate volume, biopsy results, mpMRI results, etc.—biopsy coordinates were used to compute various spatial features that were likely predictive of bilateral or midline extension of csCaP (Figure 3). Examples include but are not limited to: distance between prostate midline and nearest positive biopsy core, distance between midline and the nearest suspicious PI-RADSv2 lesion, and the presence of a negative biopsy core between a biopsy-confirmed lesion and midline. All features involving distance measures were scaled to a 40cc prostate, approximated as a sphere. A full list of features considered can be found in Supplemental Table 1.