Fig. 3
Voxel-wise classification results for nine external validation samples not used during training or testing. The columns are grouped by diagnostic category: the first three columns show healthy tissue samples, the next three represent goiter, and the final three correspond to cancerous tissue. The first row displays the H&E-stained adjacent histological slides for visual reference, and the second row shows the binary masks of the regions of interest (ROI) used in the classification. The remaining rows present the classification maps generated by the three deep learning models: FCNN, LSTM-RNN, and 1D-CNN. In the classification maps, each voxel is colored according to its predicted class: green for healthy tissue, blue for goiter, and red for cancer.
