Dark-Field and Attenuation X-ray U-Net for Lung Tumor Detection

Dark-Field and Attenuation X-ray U-Net for Lung Tumor Detection Steps / Order to Run Code The folder 2D_ATTN_DFI_WithDisplay — this archive contains the mouse attenuation (ATTN) and dark-field (DFI) projection data, along with MATLAB code for: - Lung segmentation - Synthetic tumor insertion - Patch generation - U-Net training and testing for tumor detection --- 1. (Optional) Lung segmentation Run FlipCreateMasksDataAugMouseAll.m This re-segments the left and right lungs (2 lungs × 7 mice). Note: Not recommended to re-run unless you wish to redraw the lung masks manually. --- 2. (Optional) Tumor insertion Run Tumor_insertion.m Visualize existing tumor distributions or modify parameters such as tumor density, size, contrast, and noise. --- 3. Display tumors Run DisplayTumors.m Displays currently inserted tumors without re-running the insertion step. --- 4. Patch generation Run make_patches.m Generates image patches for training and testing. To achieve approximately balanced numbers of tumor-positive and tumor-negative patches, adjust tumorDensity parameter in Tumor_insertion.m, rerun both scripts iteratively if needed. --- 5. U-Net training and testing Change directory to UNET_training_testing/ — this folder contains scripts for three training modes: (a) ATTN-only - Training: UNET_training_testingOneChAttn.m - Testing: UNET_testingOneChAttn.m (b) DFI-only - Training: UNET_training_testingOneChDfi.m - Testing: UNET_testingOneChDfi.m (c) Dual-channel (ATTN + DFI) - Training: UNET_training_testing.m - Testing: UNET_testing.m All training and testing outputs are saved automatically in: patches_32_anyOverlap_num/ --- For questions, contact: Joyoni Dey -- [email protected]