Research

Dental caries are common chronic infectious oral diseases affecting more than 90 percent of all dentate adults and more than two-thirds of children in the United States. Early Detection of carious lesions can result in the implementation of non-surgical preventive approaches to reverse the demineralization process. The conventional approach for diagnosing dental caries is clinical examination supplemented by radiographic evaluation. However, studies based on the clinical and radiographic examination methods often show low sensitivity and high specificity. Optical coherence tomography (OCT) is a noninvasive optical imaging modality based on low-coherence interferometry that utilizes non-ionizing near-infrared laser to obtain images with micrometer resolution. Currently, the major biomedical application of OCT is in ophthalmology. Many other applications of OCT are under investigation as researchers take advantage of the ability to rapidly acquire images noninvasively. Machine learning and deep learning techniques can be used to supplement OCT imaging system to more accurately identify diseased and damaged tissue. In this research, we investigate novel approaches combining OCT imaging modality and deep convolutional neural networks (CNN) for the early detection of occlusal carious lesions. In this project, we also develop signal and image processing algorithms to extract meaningful features from OCT data for image classification and analysis.

3D Digitization of dental model is growing in popularity for dental application. Classification of tooth type from single 3D point cloud model without assist of relative position among teeth is still a challenging task. In this research, 8-class posterior tooth type classification is investigated using convolutional neural network (CNN)-based occlusal surface morphology analysis. Considering the logical hierarchy of tooth categories, a hierarchical classification structure is proposed to decompose 8-class classification task into two-stage cascaded classification subtasks. Image augmentations including traditional geometrical transformation and deep convolutional generative adversarial networks (DCGANs) are applied for each subnetwork and cascaded network. Grad-cam, a visualization method for explaining the decision-making process of CNN-based models, is employed to highlight the important region where the network pays more attention to identify a specific class in the input image. The proposed method has advantages of easy training, great ability to learn discriminative features from small image region.

The accumulation of dental plaque on a tooth surface plays a crucial role in developing dental caries. In this research, fluorescence imaging modality with structured light-based intraoral 3D scanner are combined to investigate the 3D distribution of dental plaque. The traditional fluorescence imaging method only reveals the 2D spatial distribution of the dental plaque on a tooth surface. To visualize the 3D distribution of the dental plaque on an occlusal surface, we investigate mapping 2D fluorescence images to 3D occlusal surfaces. The 3D distribution of occlusal plaque reveals that dental plaque accumulation relates to the local and global morphology of the tooth surface. The investigation of the 3D distribution of occlusal plaque using 2D-3D registration paves the path for the quantitative analysis of the tooth surface morphology to perform plaque-guided caries risk assessment.

Migraines are the sixth most disabling disease in the world, affecting more than 1 billion people worldwide and are just one illness affected by environmental triggers due to changes to that occur inside the home. Due to migraines similarities to sinus headaches, migraines affected by environmental triggers can often be misdiagnosed. In this research work, an iOS-based environmental analyzer was designed, implemented and evaluated for migraine sufferers with the use of sensors. After the data collection and cleaning (data from a user’s surroundings, i.e. temperature, humidity, pressure and altitude), five machine learning model were used to estimate prediction accuracy of migraines in terms of the environment. Preliminary results demonstrate the feasibility of using machine learning algorithms to perform the automated recognition of migraine trigger areas in the environment. (Sponsored by iDevices)

Another research project is focused on novel approaches combining machine learning and digital signal processing for real-time robot arm manipulation to pave the path for developing smart prosthetic devices for amputees and innovative robotic rehabilitation systems. The Electromyography (EMG) signals are generated by human muscle systems when there are any movements and muscular activities. These signals are detected over different areas from the skin surfaces and each movement corresponds to a specific activation pattern of several muscles. In this study, multi-channel EMG measurements were performed with electrodes placed on involved arm muscles. The EMG signals were digitally recorded and processed using digital filters, feature extraction methods, and classification algorithms. 

Photoacoustic imaging is an emerging imaging modality that images optical absorption contrast in tissue based on the thermoelastic induction of acoustic signals as a result of the absorption of pulsed or modulated optical energy. To obtain photoacoustic images, an optical pulse propagates into a tissue sample diffusively and the photoacoustic signals arising from tissue thermoelastic expansion caused by the optical irradiation are measured around the sample via ultrasound transduction. In this project, we develop and investigate low-cost photoacoustic microscopy system, which is capable of mapping microvasculature networks in biological tissue and resolving blood vessels with much higher spatial resolution than conventional photoacoustic imaging with ultrasound array transducers. We also analyze photoacoustic images using machine learning, pattern recognition, and signal/image processing techniques.

Legume crops such as pea and bean develop symbiosis with a group of bacteria, called Rhizobia, and obtain their nitrogen requirement from the atmosphere. The process of biological nitrogen fixation (BNF) takes place in small nodules that are formed on the legume root system. Due to their capability for biological nitrogen fixation, legumes do not need nitrogen fertilizers, and can add to soil nitrogen for succeeding crop. Thus, legume crops are well known for their contribution to soil nitrogen through the BNF. However, the nitrogen benefits of legumes significantly affected by cultural practices and environment. Sophisticated methods such as staple isotope technique is used to estimate the nitrogen benefits of legumes. In addition, it is very common to quantify plant nitrogen fixation through counting the number of nodules on the legume root system. Nodule number, nodule mass, and nodule shape are seen to be correlated with legume’s BNF. Due to the slow process of counting nodules by hand, this valuable trait is not often included in legume research. In collaboration with College of Agriculture at California State University, Chico, we develop deep learning models and digital image processing techniques to analyze legume roots and characterize their nodulation.