Disease Classification of Chest X-ray using CLIP and Few-shot Learning

MIMIC-CXR 데이터셋과 CLIP을 이용한 구조화된 임상적 소견 예측

Dataset

Method

FlexR (Few-shot classification with Language Embeddings for chest X-ray Reporting)

1단계. CLIP (Contrastive Language-Image Pretraining)

encoder	model
image encoder	ResNet50
text encoder	Transformer

2단계. Language embeddings of clinical findings

3단계. Fine-tuning the classifier

Demo

https://cxr-fsc.streamlit.app/

Teammate - R&R

Deep Daiv. M3


정수미	김예린	홍재령
1단계	2단계	3단계