Chaoyi Wu

I am a PhD candidate working on medical image analysis and machine learning at Shanghai Jiao Tong University.

My current research interest is in building multimodal fundation models for medicine.

I am currently looking for an opportunity for a post-doctoral position and would be very happy to talk to you further.

In this study, we present MedS-Bench, a comprehensive benchmark designed to evaluate the performance of large language models (LLMs) in clinical contexts beyond multiple-choice question-answering. Moreover, we build up a new comprehensive medical instruction dataset, termed as MedS-Ins.

In this paper, we propose a grounded report generation system for brain MRI leveraging the coorperation of different sub-tools. In real clinical scenerios, the system can significantly improve the efficiency of the radiologists.

In this paper, we propose an entity-level assessment metric for radiological reports beyond chest x-ray using NER and synonym normalization models. Unlike LLM-based assessment pipelines, our metric is more lightweight and objective targeting large-scale auto-evaluation.

In this paper, we introduce RadGenome-Chest CT, a comprehensive, large-scale, region-guided 3D chest CT interpretation dataset based on CT-RATE. It includes: Organ-level segmentation for 197 categories; 665K multi-granularity grounded reports; 1.3M grounded VQA pairs.

In this paper, we consider the problem of visual representation learning for computational pathology, by exploiting large-scale image-text pairs gathered from public resources, along with the domain specific knowledge in pathology.

In this paper, we aim to develop a multilingual language corpus (MMedC), benchmark (MMedBench) and an open-source, multilingual language model (MMedLM) for medicine, that benefits a wider, linguistically diverse audience from different regions.

In this paper, we build up a universal medical segmentation model, driven by text prompts (SAT).

In this paper, we collect a large-scale multi-modal, multi-scan, long-tailed muti-lable diagnosis (classification) dataset. We further propose a vision encoder together with a fusion module, enabling arbitrary scan input per case. On evaluation, our methods achieve better experiment results on our benchmark and can also serve as a pre-train model for external datasets.

We evaluate the GPT-4V on 92 radiographic cases, 20 pathoglogy cases and 16 location cases across 17 medical systems covering 8 imaging modalities. In general, as the cases shown, GPT-4V is still far from clinical usage.

We release a new knowledge-enhanced Brain MRI pre-train foundation model leveraging image-report pairs which can realize zero-shot diagnosis of unseen brain diseases.

In this study, we aim to initiate the development of Radiology Foundation Model, termed as RadFM. we construct a large-scale Medical Multi-modal Dataset, MedMD, consisting of 16M 2D and 3D medical scans.

In this paper, we focus on the problem of Medical Visual Question Answering (MedVQA). We propose a generative medical VQA model, MedVInT, together with a large scale MedVQA Dataset, PMC-VQA.

In this report, we introduce PMC-LLaMA, an open-source language model that is acquired leveraging large medical corpus, surpassing chatGPT on medicalQA benchmarks.

We collect a biomedical dataset, PMC-OA with 1.6M image-caption pairs collected from PubMedCentral's OpenAccess subset.

Here, we propose a knowledge-enhanced vision-language pre-training approach for auto-diagnosis on chest X-ray images. First trains a knowledge encoder based on an existing medical knowledge graph, then leverages the pre-trained knowledge encoder to guide the visual representation learning.

In this paper, we consider the problem of disease diagnosis. Unlike the conventional learning paradigm that treats labels independently, we propose a knowledge-enhanced framework, that enables training visual representation with the guidance of medical domain knowledge.

We propose to leverage medical specific knowledge enhancing language-image pre-training method, significantly advancing the ability of pre-trained models to handle unseen diseases on zero-shot classification and grounding tasks.

We propose Boundary-Enhanced Self-SupervisedLearning (BE-SSL), leveraging supervoxel segmentation and registrationas two related proxy tasks, enhancing brain structure segmentation.

We first leverage the information of LN stations for metastatic LN detection. Metastatic LN station classification is proposed as proxy task for metastatic LN detection. A GCN-based structure is adopted to model the mutual influence among LN stations.

Based on a template by Jon Barron.