Phd thesis: 3D-2D Representation Learning for Longitudinal Analysis using Single-slice CT Scans• Design a Hierarchical transformer target to include the local positional information to segmentation large number of tissues (e.g. 133 brain tissues) and hierarchical structures.• Design a conditional VAEGAN-based model to generate a target slice at a pre-defined vertebral level given arbitrary abdominal slice to allow longitudinal single slice CT body composition analysis.• Design a 3D-to-2D knowledge distillation framework to improve 2D model segmentation with 3Dmodel trained with volumetric data.• Characterize the correlation between body composition change on aging and cognitive decline.

Camera key points detection and calibration• Generate fake DSA images using conditional diffusion model to solve data scarcity problem• Detect bifurcation points in the DSA images and match key points from different views using GraphNeural Network.

Click XY Heatmap Spaminess Prediction and User Agent Stats Analysis• Designed a multi-modal algorithm to incorporate image and text information for spaminess prediction to catch currently missed spam.• Calculate user agent related information to capture additional spam.

Prostate Cancer Detection using Multi-parametric MRI images• Designed a panoptic segmentation method fusing the information from different modalities as well as utilizing the global and local information for prostate cancer detection• Designed a false positive reduction network to increase the detection specificity by employing multiscale information, multitask loss function and attention mechanism to learn better feature representation

Real-time Assessment of Catheter Contact Quality using OCT Images• Proposed a novel way to provide real-time guidance during radiofrequency ablation (RFA) procedure by analyzing the contact quality between the RFA catheter and cardiac tissue using M-mode OCT images.• Designed an algorithm for catheter contact and orientation assessment and verified its feasibility by conducting in vitro experiments.