How soribido/bmd_prediction works
Project Type and Purpose
• This is a research/experimental project centered on medical imaging analysis. It leverages deep learning for segmentation and regression (BMD estimation), alongside an XAI component for model interpretability.
• The README provides a paper reference, experimental cases (cropped images, entire vertebrae-masked images, vertebral-body-masked images), and details on various workflows including segmentation and prediction, which indicate a clinical/medical image processing research purpose.
File Structure
• Top-Level Files: Standard files such as .gitignore, LICENSE, and README indicate proper project documentation and versioning.
• Directories and their Purposes:
• XAI: Contains a notebook (XAI.ipynb) likely showing the implementation/demonstration of the modified Grad-CAM (Grad-RAM and Grad-RAMP) methods for model interpretability.
• bmd_estimation: Houses notebooks (bmd_CA_case1.ipynb, bmd_pred_case1.ipynb) that likely demonstrate the process and experiments involving BMD prediction.
• bone-segmentation: Contains the segmentation notebook (seg_body_model.ipynb), which uses a U-Net based model to segment the L1 region.
• preprocessing: Includes Python scripts (check_ptinfo.py, ctdicom_check.py, preprocessing_image.py) responsible for preprocessing CT images and ensuring data quality/preparation.
• Clear labels for each primary component:
• "Preprocessing Module" (with sub-tasks like DICOM checks and image formatting)
• "Segmentation Module" (U-Net based segmentation of L1 region)
• "BMD Prediction Module" (Deep residual CNN for BMD estimation)
• "XAI Interpretation Module" (Modified Grad-CAM techniques - Grad-RAM/Grad-RAMP)
Raw CT Images → Preprocessing → Segmentation → Prediction → XAI.