U-Net Guided Model (Humeral Head ROI)
This model performs an anatomical preprocessing step before classification.
A U-Net segmentation network first isolates the humerus and
automatically crops the image around the humeral head, which is
the region where rotator cuff calcifications most frequently appear.
-
Probability Score: Estimates the likelihood of calcific tendinopathy.
-
Visual Heatmap: Highlights the image regions that most influenced the prediction.
-
Overlay Visualization: Displays the prediction results directly on the image for easier interpretation.
Clinical decision-support tool only — not intended as a standalone diagnosis.
Select Model
This model combines semantic segmentation and classification. A dedicated U-Net segments the humerus and is used to center the image on the humeral head, reducing background noise and focusing analysis on the most relevant anatomy for detecting rotator cuff calcific deposits.
A DICOM preprocessing pipeline extracts raw pixel data, adjusts contrast using Window Center/Width, and normalizes intensities to ensure consistent visualization. The pipeline then applies targeted cropping, padding, and resizing to standardize inputs while preserving the humeral head region.
Results are displayed as a probability score plus a heatmap (Grad-CAM) and an image overlay, helping users understand where the model is “looking” when estimating the likelihood of calcific tendinopathy.