Weicheng Dai

I am currently a PhD student at Boston University, where I am very fortunately advised by Professor Kayhan Batmanghelich. Previously I worked at Yale University for Professor Julius Chapiro and Professor James S. Duncan. I was also working closely with Professor Chenyu You. I obtained a Master degree in Computer Science at New York University. My research interests include computer & medical vision, and explainable machine learning.

Email  /  LinkedIn  /  Github  /  Google Scholar

In this work, we propose MedSynV2, a flexible multimodal framework for controllable volumetric image generation that supports input from radiology reports and segmentation prompts (both optional). Our approach allows users to provide segmentation of a specific anatomy or abnormality without requiring full-organ annotations. The semantic meaning of the segmentation mask is specified through an accompanying text description, resulting in a highly flexible and scalable conditioning mechanism.

PhyDiCT is a training-free, plug-and-play framework for reconstructing 3D lung CT volumes from sparse X-ray projections. It combines a differentiable physics-based forward model grounded in the Beer-Lambert law, a frozen text-conditioned diffusion model as a strong 3D CT prior, and split Gibbs sampling to jointly enforce projection fidelity and prior consistency. Without any task-specific training or fine-tuning, PhyDiCT outperforms fully trained CT reconstruction methods, achieving up to +7.5% SSIM on public 3D CT benchmarks.

In this work, we propose CFR, which focuses on exploring mitigating the reliance on spurious features for CLIP without using any group annotation. We showcase a lightweight representation calibration method for fine-tuning CLIP, by first generating a calibration set using the pretrained CLIP, and then calibrating representations of samples within this set through contrastive learning.

Two practical solutions via stratified group sampling theory that correct for the variance introduced by the common sampling practice, and achieve significant performance benefits.

(* denotes equal contribution) Three simple principles: (1) tailness: giving more importance to tail class hard pixels; (2) consistency: enforcing the feature invariances to specified data transformations; (3) diversity: ensuring anatomical diversity in the set of different sampled images in those imbalanced, unlabeled, and diverse scenarios.

In this work, we present ACTION, an Anatomical-aware ConTrastive dIstillatiON framework, for semi-supervised medical image segmentation.

In this work, we propose MORSE, a generic implicit neural rendering framework designed at an anatomical level to assist learning in medical image segmentation. The core of our approach is to formulate medical image segmentation as a rendering problem in an end-to-end manner.

In this work, we present ACTION++, an improved contrastive learning framework with adaptive anatomical contrast for semi-supervised medical segmentation. We propose an adaptive supervised contrastive loss, where we compute the optimal locations of class centers uniformly distributed on the embedding space. We also use dynamic Tau to yield better separation between majority and minority classe.