Algorithm 1 Steps for predicting lesions with extraprostatic extension

Require:${\mathcal{I}}_{T2},{\mathcal{I}}_{ADC},{\mathcal{L}}_{pr}$ Require: user defined parameters $\alpha ,TC{L}_{th}$ Ensure: load pre-trained cancer model $f_{\theta }$     $p_{Ca}\leftarrow f_{\theta }({\mathcal{I}}_{T2},{\mathcal{I}}_{ADC})$              ▹ deep learning model inference, Equation (1)   ${\mathcal{M}}_{pr}\leftarrow$ dilate$\left({\mathcal{L}}_{pr}\right)$                                  ▹ Equation (2)   $p\leftarrow p_{Ca}\ast {\mathcal{M}}_{pr}$                         ▹ apply dilated prostate mask, Equation (3)   $p_{\alpha }\leftarrow$ binary threshold $(p;\alpha )$                            ▹ Equation (4)   ${\mathcal{C}}_{\alpha }\leftarrow$ connected components $\left(p_{\alpha }\right)$                         ▹ Equation (5)   $n\leftarrow len\left({\mathcal{C}}_{\alpha }\right)$     $q\leftarrow$ empty list                ▹ where final EPE lesion predictions will be appended   for i in $0,1,...n-1$do                       ▹ iterate over all candidates ${\mathcal{C}}_{\alpha }^{\left(i\right)}$     if ${\mathcal{C}}_{\alpha }^{\left(i\right)}\cap {\mathcal{L}}_{pr}\ne 0$ and ${\mathcal{C}}_{\alpha }^{\left(i\right)}\setminus {\mathcal{L}}_{pr}\ne 0$ then               ▹ Equations (6) and (7)         $l_{TCL}^{\left(i\right)}\leftarrow$ compute tumor–capsule contact line length$({\mathcal{C}}_{\alpha }^{\left(i\right)},{\mathcal{L}}_{pr})$         ▹ Equation (8)         if $l_{TCL}^{\left(i\right)}>TC{L}_{th}$ then            $q\leftarrow \mathrm{append}({\mathcal{C}}_{\alpha }^{\left(i\right)}\ast p_{\alpha })$         ▹ component ${\mathcal{C}}_{\alpha }^{\left(i\right)}$ satisfied all criteria, Equation (9)                                      ▹ append to list of final EPE predictions         end if     else continue     end if end for $Q\leftarrow$ aggregate all final predictions for the case, q