Algorithm 4 Multimodal feature preparation

Require:  Radiomics feature matrix R, deep feature matrix D, target labels y Ensure:  Preprocessed and fused feature matrix F    1:Apply KNN imputation with $k=5$ to R, producing $R_{imp}$    2:Apply KNN imputation with $k=5$ to D, producing $D_{imp}$    3:Standardize $R_{imp}$ with zero mean and unit variance, yielding $R_{norm}$    4:Standardize $D_{imp}$ with zero mean and unit variance, yielding $D_{norm}$    5:Rename columns of $D_{norm}$ as deep_feature_i for traceability    6:Concatenate $R_{norm}$ and $D_{norm}$ horizontally to form F    7:Align F rows to patient IDs from y    8:return F