Instance Segmentation and Teeth Classification in Panoramic X-rays

This repository contains the implementation and dataset related to the paper Instance Segmentation and Teeth Classification in Panoramic X-rays, submitted to Expert Systems with Applications Journal.

Dataset

• We introduce a set of 425 panoramic X-rays with Human annotated Bounding Boxes and Polygons, the 425 images are a subset of UFBA-UESC Dental Dataset. This dataset can be extensively used for detection and segmentation tasks for Dental Panoramic X-rays. Refer to Description for understanding the organisation of annotations and panoramic X-rays. The Distribution of Categories in the dataset are metnioned in the table below.

Category	32 Teeth	Restoration	Dental Appliance	Images	Used Images
1	✓	✓	✓	73	24
2	✓	✓		220	72
3	✓			45	15
4	✓			140	32
5	Images containing dental implant			120	37
6	Images containing more than 32 teeth			170	30
7		✓	✓	115	33
8		✓		457	140
9			✓	45	7
10				115	35
Total				1500	425

Results

• Teeth Numbering Results

Model Architecture	mAP	AP50
Mask R-CNN	70.5	97.2
Mask R-CNN + FCN	74.1	92.8
Mask R-CNN + pointRend	75.3	94.4
PANet	74.0	99.7
HTC	71.1	97.3
ResNeSt	72.1	96.8
YOLOv8	72.9	94.6

• Instance Segmentation Results

Model Architecture	Incisors	Canines	Premolars	Molars
U-Net	73.29	69.92	67.62	64.98
Mask R-CNN	89.56	89.45	88.70	87.55
U-Net + Mask R-CNN	91.55	91.00	90.00	88.58
BB-UNet + YOLOv8 ( Test Dataset 1)	85.81	84.91	84.89	84.40
BB-UNet + YOLOv8 ( Test Dataset 2)	85.71	86.64	86.22	86.03

• Refer to the paper for further information on model architectures and datasets used for evaluation.

Teeth Numbering Heatmaps

Segmentation Masks

Code Structure

2ddaatagen.ipynb                   => Notebook for generating labels
yolov8_train.ipynb                 => Notebook for training YOLOv8
yolo_test.ipynb                    => Notebook for testing YOLOv8
unet_training.ipynb                => Notebook for training U-Net
unet+cv.ipynb                      => Notebook for training U-Net with cross validation
yolov8+unet_training.ipynb         => Notebook for training BB-UNet
yolov8+unet+cv.ipynb               => Notebook for training BB-UNet with cross validation

Cite Us

Cite the paper if you find our work useful.

@misc{budagam2024instance,
      title={Instance Segmentation and Teeth Classification in Panoramic X-rays}, 
      author={Devichand Budagam and Ayush Kumar and Sayan Ghosh and Anuj Shrivastav and Azamat Zhanatuly Imanbayev and Iskander Rafailovich Akhmetov and Dmitrii Kaplun and Sergey Antonov and Artem Rychenkov and Gleb Cyganov and Aleksandr Sinitca},
      year={2024},
      eprint={2406.03747},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}