This repository contains the machine learning models developed for dental image classification and detection. The goal of the project is to train AI models to detect and classify various dental conditions, including oral lesions, cavities, fillings, and plaque, using advanced image processing techniques.
The project consists of two main components:
- Oral Lesions Project: Using CNN for image classification to detect eight different types of oral lesions.
- Oral Condition Project: Implementing YOLOv5 for object detection of dental conditions (plaque, fillings, and cavities) and using UNet for semantic segmentation to assess the severity of the conditions.
- Image Classification: Common CNN model for detecting oral lesions with over 94% accuracy.
- Object Detection: Fine-tuning a YOLOv5 model to detect dental conditions in images.
- Semantic Segmentation: Using UNet to segment oral conditions for further analysis.
- CNN (Convolutional Neural Network): Used for classifying oral lesions.
- YOLOv5 (You Only Look Once): Applied for object detection to identify various oral conditions.
- UNet: Used for semantic segmentation to evaluate the severity of detected conditions.
src/
: Contains the source code for model training and evaluation.data/
: Placeholder for the dental image datasets (not included due to size constraints).notebooks/
: Jupyter notebooks for training models and conducting experiments.data pipeline/
: Preprocessing data and data visualization
Two main datasets are used in this project:
- Oral Lesions Dataset: Contains annotated images of oral lesions.
- Oral Conditions Dataset: Includes images with annotations for plaque, cavities, and fillings.
- Clone the repository:
git clone https://github.com/SLAM-CROC/AI-innovate-Dental-Imaging.git
- Install the required dependencies:
pip install -r requirements.txt
- Download the datasets and place them in the
data/
folder. - Train the models using the provided notebooks:
oral_lesions_colab.ipynb
oral condition detection.ipynb
- View data preprocessing and data visualization in the
data pipeline/
folder.
- Oral Lesion Classification: Achieved ~94% accuracy on the testing set.
- Object Detection (YOLOv5): Successfully detected dental conditions with good precision and recall.
- Semantic Segmentation (UNet): Segmented images with varying levels of success, limited by the dataset size and manual labeling.