1. First, you are adviced to train the segmentor module using codes under ./joint_model with the combined samples from both domains, and save the best model of lesion segmentor. Then, train the lesion segmentor and maligancy classifier modules simultaneously (initialized with the best weight of the lesion segmentor) to accelerate model convergence , and save the best PLDC models (i.e., T2: weight2_1.h5, ADC: weight3_1.h5, hDWI: weight4_1.h5).