11/14/2023
By Xizhe Wang

The Kennedy College of Sciences, Miner School of Computer & Information Sciences, announces the doctoral dissertation defense of Xizhe Wang entitled: "Deep Learning for Dental Caries Detection with Model Enhancement and Dental Dataset Expansion."

Ph.D. Candidate: Xizhe Wang
Date: Monday, Nov. 20, 2023
Time: 9:30 a.m. Eastern Time
Location: This will be a virtual defense via Zoom

Committee Members:


  • Yu Cao (advisor), Professor, Director, UMass Center for Digital Health (CDH), Miner School of Computer & Information Sciences
  • Benyuan Liu (advisor), Professor, Director, Miner School of Computer & Information Sciences
  • Hengyong Yu (member), Professor, Department of Electrical & Computer Engineering

Abstract:

Deep learning has witnessed outstanding progress in recent decades. The Deep Neural Network (DNN), serving as an essential constituent of deep learning, focuses on artificial neural networks with multiple layers. This architectural design enables DNNs to automatically learn and extract high-level representations from raw data. Moreover, the development of unsupervised and semi-supervised machine learning pipelines has contributed to the improvement of deep learning applications through dataset enhancement techniques. These advancements have led to breakthroughs across diverse domains, including Computer Vision (CV), Natural Language Processing (NLP), and Automatic Speech Recognition (ASR). This study aims to investigate and develop deep learning-based approaches for the detection of dental caries. The research will be accomplished through the advancements in neural network architectures and enrichment of dental datasets.

In the medical and healthcare sectors, deep learning approaches have demonstrated remarkable efficacy and efficiency. Dental caries, as the most prevalent oral disease, presents a significant healthcare challenge. It affects more than half of the global population and significantly diminishes individuals' quality of life by impairing their eating and socializing abilities. Consistent dental check-ups and professional oral healthcare play a crucial role in preventing dental caries and other oral diseases. Deep learning-based object detection provides an efficient approach to assist dentists in identifying and treating dental caries, thereby contributing to enhanced oral health outcomes.

DNN-based object detection techniques offer promising solutions to improve the efficiency of dental caries diagnosis. In our dental dataset, there are 1,008 panoramic dental X-ray images, with each image having a resolution of 2,918 × 1,435 pixels. The dataset includes a total of 31,296 annotations of teeth, categorized into ten distinct dental classes and represented by both bounding box and segmentation labels. The Cascade Mask R-CNN is selected as the baseline model due to its ability to utilize both types of labels effectively. In addition, it is observed that teeth are concentrating within a Region of Interest (RoI) in each dental X-ray image due to the anatomical characteristic. In this paper, we present a deep learning framework with a lightweight pruning region of interest (P-RoI) proposal specifically designed for detecting dental caries in panoramic dental radiographic images. Moreover, this framework can be enhanced with an auxiliary head for label assignment during the training process.

Furthermore, it is widely acknowledged that the performance of DNN models heavily relies on the availability of sufficient and accurately labeled data. However, the collection and annotation of dental X-ray images encounter obstacles due to privacy concerns and the requirement for specialized expertise. Consequently, the limited access to labeled dental image datasets restricts the potential of DNNs in supporting oral and dental healthcare. Self-Training (ST) is a semi-supervised machine leaning approach that addresses this problem to a large extent. It repeats the procedures of training a model on the labeled dataset, and then applying it to generate pseudo labels on the unlabeled dataset, and further using the combined data with the original and pseudo labels to train new models. Nevertheless, the latent errors of the pseudo labels can arise and even be amplified throughout the ST pipeline, which leads to a significant performance decline for DNN models. In this paper, we also propose a Greedy algorithm-based Self-Training (Greedy-ST) pipeline to address this problem. At each iteration, the Greedy-ST selects an optimal confidence threshold to generate predictions as pseudo labels, and uses static fine-tuning (SFT) and dynamic fine-tuning (DFT) to refine them.