Algorithm 1: hotBSI algorithm

Inputs:   NM - feature set from all the hotspots extracted from the nonmalign images   M - feature set from all the hotspots extracted from the malign images   T - threshold (default as 0.8)   NrIt - number of iterations (default as 100) Output:   C - a classifier to classify new hotspots as nonmalign or malign  1:Train an initial classifier, $C_0$, with the input features (NM ∪ M)  2:for i = 1:NrIt do  3:    Empty M  4:    for each patient in the malign set do  5:        Use $C_{i-1}$ to predict the probabilities of the detections to be a metastases ($P_{\mathrm{met}}$)  6:        Identify the hotspot with the highest likelihood of being a metastasis ($P_{\max }$)  7:        for d = 1: number of detected hotspots for the current patient do  8:           if $P_{\mathrm{met}}\left(d\right)==P_{\max }$ || $P_{\mathrm{met}}\left(d\right)>T$ then  9:               Add the hotspot to M 10:    Create a new training set, NM ∪ M 11:    Train a new classifier $C_i$ with the new training data set 12:return$C_{NrIt}$