CC BY
7DETECTION AND DIFFERENTIAL TREATMENT OF PATHOLOGIES IN X-RAY
DENTAL IMAGES 1Ismailov Otabek, 2Iskandarova Sayora, 3Temirova Xosiyat Farxod qizi
1TATU professor, 2TATU assistant of professor, 3TATU doctoral student https://doi.org/10.5281/zenodo.10721391
Abstract. Artificial Intelligence (AI) is gaining traction in medical imaging. There are a plethora of potential applications, covering every stage of the medical imaging life cycle from image creation to diagnosis to outcome prognosis. The use of Artificial Intelligence (AI) and Machine Learning is having a rising influence in the field of dentistry and is improving the growth of digital tools and technology, with a wide range of applications in cosmetic dental treatments and treatment planning. Deep learning has been able to match and improve human performance in fields such as image processing for extremely complex tasks such as object detection models with classification and prediction capabilities, as well as the identification of numerous dental diseases and their differential treatments. The lack of sufficiently large, curated, and representative training data with expert labeling (eg, annotations) is one of the main barriers to the development and clinical application of AI algorithms. This pilot study used a dataset of 664 dental panoramic X-ray images to build a model that can detect dental diseases and the differential treatments present on a full dental X-ray. After being manually labeled for 9 different classes.
Keywords: X-ray films, Machine Learning, diagnosis, performance, algorithms, differentia, dental panoramic.
The state of dentistry has improved recently, as has the number of persons with dental issues. Dentists are frequently required to treat a significant number of patients in a single day, and a substantial number of dental X-ray [1] films are taken every day as an important diagnostic tool to aid dentists. The majority of dentists interpret X-ray films, which takes up important clinical time and increases the risk of misdiagnosis or under diagnosis owing to individual factors like as exhaustion, emotions, and inadequate skill levels. Consequently, relying solely on dentists may occasionally impede treatment. Therefore, by reducing dentists' workload and the likelihood of incorrect diagnoses, intuitive dental disease detection systems [2] may raise the standard of dental treatment.
A panoramic X-ray is a two-dimensional X-ray that captures a picture of the entire mouth in a single image. The X-ray shows a view of the entire set of teeth, various bones in the head and neck, and other important anatomical components [3]. Unlike bitewing dental X-rays, which show a close-up of your teeth, the panoramic X-ray provides doctors with a comprehensive image of a patient's head and neck. This perspective enables the diagnosis of more than simply routine dental issues such as cavities or gum disease, rather, this perspective allows clinicians to notice other crucial concerns in the surrounding tissue and jaw bones, such as oral cancer or other abnormalities, that would not be evident on a bitewing X-ray. Deep learning a subset of artificial intelligence (AI) can be applied to real-world problems and is used in many areas of society, including the medical and dental fields. Convolutional Neural Networks (CNNs), which are deep learning structures capable of extracting various features from abstracted layers of filters, are commonly used for evaluating large and complex images. Object detection architectures can categorize, detect, and perceive the desired contents from an image at a much faster rate than typical CNNs by using a couple of bounding boxes for object edge detection. In contrast to image classification, which just determines the contents of an object within an image while not specifying
the location of an object within an image, these object detection architectures determine the location of the object within the image. This work employed the deep learning-based object detection technique YOLO to train models in order to detect and classify various dental issues and treatments. Prior to YOLO, the field was dominated by the two-stage object detection model. It uses region-based classifiers to find regions before passing those locations along to a more potent classifier. Despite the fact that this approach yields accurate results with a high mean Average Precision (mAP), it is highly resource-intensive and calls for several iterations. YOLO proposed an alternative methodology in which both steps are carried out in the same neural network. The image is initially partitioned into cells, each of which has an identical S x S dimensional segment. Each cell then recognizes and locates the objects it contains using bounding box coordinates using the object label and probability that the object is present in the cell [5]. The YOLO v6 [6] model is used in this work to identify the most common radiographic findings, such as root canal treatment, dental caries, restorations, dental implants, orthodontic operations, crowns, bridges, and root stumps. Label Img will be used to segment our images, build bounding boxes, and generate coordinates for each training image. The dataset's annotations are divided into nine classes: impactions, dental caries, root stumps, root canal treatment, implants, restorations, crowns, bridges, and orthodontic treatment. The annotations for each image are recorded in the same folder in a'.txt' file with the appropriate name and in the YOLO labeling format. The ".txt" file that was saved contains annotations for the corresponding image file, such as the object class, object
Fig. 1. Overall data flow pipeline:
First, the dataset is collected and labeled using an open access tool, then split into three parts, normally into 70% of training, 20% of validation set and 10% of test set. After that, DL models are trained from scratch, and their training loss plots are obtained to indicate the significance of the models. Then, performance metrics are used for the classification of images, and finally, visualization techniques are used to detect, localize and classify the images.
Image classification using a hybrid neural network CNN-LSTM-ELM is a hybrid neural network consisting of a combination of convolution operation, LSTM (Long Short-Term Memory), ELM (Extreme Learning Machine) classifier. Tooth A CNN-LSTM-ELM algorithm was proposed for image recognition (HAR-Human activity recognition). used an image database
of 3495 training samples to evaluate the hybrid neural network architecture. Figure 1 shows the general architecture of the CNN-LSTM-ELM model.
Fig. 2 . CNN-LSTM-ELM model.
Methodology The object detection task is to recognize the existence of objects in an image and to provide a bounding box indicating their specific location as well as the labeled classes of the exact objects as output. The goal was not to create software that would entirely replace doctors, but rather to create a proof-of-concept model that would provide a second opinion and make reporting easier for doctors. The annotated panoramic images serve as the training dataset for the model. Fig. 1. depicts the overall work flow of this work, where the The dataset is collected and labeled using an open-access tool, then divided into three sections, typically 70% training, 20% validation, and 10% test. Following that, Deep Learning models are trained from scratch, and training loss plots are created to determine the models' significance. The images are next classified using performance measures, and ultimately, visualization techniques are employed to detect, localize, and classify the images, finally generating the output images with object detection creating bounding box and their corresponding confidence score. This research presents a method for using cutting-edge YOLO in the automated detection of dental problems and their differential treatments. YOLO is a visually appealing CNN-based system for image and video object detection, classification, and localization. YOLO is a well-known object detection algorithm because of its speed and accuracy. With the birth of YOLO v6, it achieves the best trade-off in terms of accuracy and speed so far. YOLO pioneered the usage of a separate neural network for the entire image. This network splits the image into regions and predicts bounding boxes and probabilities for each. Finally, these bounding boxes are weighted by the predicted probability.
Fig. 3. YOLO v6 overall architecture
The Decoupled Head structure was implemented in YOLO v6 to simplify its design. The original YOLO v5 detection head is accomplished via the fusion and sharing of classification and regression branches, but the detection head of YOLO X decouples the classification and regression branches and adds two extra 3x3 convolutional layers. To develop a more efficient decoupled head, we use a hybrid-channel technique in YOLO v6.
TABLE I Class Labels
Object class Labels Number of images
0 root canal treatment 257
1 caries 27 Ü
2 restoration 244
3 crown 147
4 impacted tooth 227
5 root stump 85
6 bridge 73
7 orthodontic treatment 42
8 implant 6
REFERENCES
1. Kumar, S., Rani, S., & Singh, R. (2021). Automated recognition of dental caries using K-Means and PCA based algorithm
2. Ismailov O. Iskandarova S. Temirova X. Algorithms for diagnosing dental diseases. IV International Scientific and Practical Conference "Innovative research and perspectives of the
development of science and technology" Stockholm, Sweden, January 29-31, 2024. 332-336 p.
3. Atencio, Y. P., Marin, J. H., Holguin, E. H., Yanqui, F. T., & Cabrera, M. I. (2021, October). Image Processing Techniques for Medical Applications. In 2021 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME) (pp. 1-6). IEEE.
4. Thulaseedharan, A., & PS, L. P. (2023, May). Detection of typical Pathologies and Differential Treatments in Dental Panoramic X-Rays based on Deep Convolutional Neural Network. In 2023 International Conference on Control, Communication and Computing (ICCC) (pp. 1-6). IEEE.
5. Goswami, M., Maheshwari, M., Baruah, P. D., Singh, A., & Gupta, R. (2021, September). Automated Detection of Oral Cancer and Dental Caries Using Convolutional Neural Network. In 2021 9th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO) (pp. 1-5). IEEE.
6. Chin, C. L., Lin, J. W., Wei, C. S., & Hsu, M. C. (2019, November). Dentition labeling and root canal recognition using ganand rule-based system. In 2019 International Conference on Technologies and Applications of Artificial Intelligence (TAAI) (pp. 1-6). IEEE.
7. Eshmuradov D., Ismailov O., Magrupova M. METHODS AND MEANS OF DIGITAL PROCESSING OF BIOELECTRIC SIGNALS //Science and innovation. - 2023. - T. 2. - №. A2. - C. 84-88.
8. O.Ismailov, M.Fozilova, S.Mirzakhalilov, M.Ismoilov. "Primary diagnosis of cardiovascular diseases using artificial intelligence methods." International scientific and practical
conference "BIG DATA, artificial intelligence and e-commerce in the development of the digital economy: problems, challenges, solutions." Tashkent, 2023, 162 - 166 p.
