Organ Segmentation and Treatment Target Delineation
Accurate delineation of tumors and sensitive structures is important for many medical applications. One example is treatment planning in cancer radiotherapy. Although many auto-segmentation algorithms have been developed and implemented in clinical practice, none of them are satisfactory and manual contouring is often required. Such failure is mainly because conventional segmentation methods are purely based on local information in the images. To solve this problem, we have been investigating deep learning-based strategies.
AI Treatment Planner
The effective radiotherapy modalities employed today — such as intensity-modulated arc therapy (IMRT) and volumetric modulate arc therapy (VMAT) — come at the cost of increased sophistication in treatment planning. The process is increasingly more time-consuming as the human has to make more trial-and-error guesses in planning parameters in order to achieve an acceptable dosimetry and deliverability. We aim to design an AI treatment planner capable of learning radiotherapy planning, improving itself by continually developing new plans, and eventually outperforming human planners while maintaining individual physician preferences.
Treatment Outcome and Toxicity Prediction
Predicting treatment outcome and toxicity is critical to select personalized options for an individual cancer patient. Oncologic outcome data is often imbalanced, thus conventional algorithms based on a single objective such as accuracy during model construction may lead to low sensitivity or specificity. To overcome the limitation of the current single objective-based predictive models, we strive to develop a multi-objective model that explicitly considers both sensitivity and specificity during model optimization, capable of obtaining a solution that balances sensitivity and specificity.
Human Error Detection and Prevention
After heart disease and cancer, medical errors are the third leading cause of death in the U.S. Many quality assurance and error detection processes are done manually by humans. In order to develop a comprehensive error checking solution, we explore deep learning methods. Our goal is to develop an AI-based error detection system (SafetyNet) that can reside in any electronic medical record or treatment management system to automatically detect any medical errors.
Image Reconstruction and Processing
We are actively exploring deep learning techniques for solving various image reconstruction and processing problems. Currently, our team is working on developing new deep learning-based methods for automatic and more accurate image reconstruction, MRI to CT conversion, CBCT to CT conversion, 4-D CT to nuclear medicine image conversion, and MRS super-resolution.
Wearable Sensors and Smart Clinic
Recent advances in telecommunications, microelectronics, and sensor-manufacturing techniques — especially with miniature circuits, microcontroller functions, front-end amplification, and wireless data transmission — have created substantial possibilities for using wearable technology for biometric monitoring. These physical, chemical, and biological sensors allow for continuous recording of health data, allowing for powerful analytics to monitor and predict patient condition, activity, location, etc. Wearable sensors will generate big data, providing the opportunity for AI to revolutionize many medical fields, such as assisted care, management of chronic diseases, as well as the development of a so-called smart clinic.