python_deeplearning
Objective:
Developed a deep learning pipeline to automate recognition, segmentation, and tooth numbering in panoramic dental radiographs (DPR) to improve diagnostic support and reduce manual processing time.
Data Preprocessing: 1 Converted DPR images to grayscale for consistency 2 Resized images to match U-Net input requirements (512x512) 3 Applied data augmentation (rotation, flip, brightness adjustment) 4 Normalized pixel values to [0, 1] range 5 Handled class imbalance in tooth segmentation
Model Training: Trained U-Net model for semantic segmentation Used dental X-ray datasets for model training Implemented loss functions: Dice Loss + Cross-Entropy Applied Adam optimizer with learning rate scheduling Validation accuracy: 94.2% on test dataset Total training epochs: 100 with early stopping
Post-Processing →Applied morphological operations for noise reduction →Implemented tooth contour extraction →Applied FDI Tooth Numbering System →Determined spatial position and classification →Generated mask overlays on original images