Chest x-ray image caption generation

In this repository we have provided the model for captioning chest x-ray images to facilitate examination of these images. Colons can be used to align columns.

Problme	X-Ray image captioning
Dataset	Indiana university chest x-ray
Dataset link	https://www.kaggle.com/datasets/raddar/chest-xrays-indiana-university
Train set	2559 pairs of frontal chest x-ray and corresponding captions
Validation set	320 pairs of frontal chest x-ray and corresponding captions
Test set	320 pairs of frontal chest x-ray and corresponding captions
Models	1. CNN-LSTM 2. Blip(https://huggingface.co/Salesforce/blip-image-captioning-base) 3. GIT(https://huggingface.co/microsoft/git-large) 4. Blip2(future work)
Evaluation metric	BLEU score

1️⃣ CNN-LSTM

Our base model is based on encoder-decoder architecture. In the image encoder part, we will use pre-trained CheXnet which is a DenseNet with 121 layers and has been trained on 112,000 chest X-ray images. To obtain text embeddings, we have used the Glove model.

2️⃣ Blip

This model has used ViT-L/16 for the image encoder part and Bert model for the text encoder part. Pretrained on COCO dataset.

3️⃣ GIT

GIT is a Transformer decoder conditioned on both CLIP image tokens and text tokens. The model is trained using "teacher forcing" on a lot of (image, text) pairs. The goal for the model is simply to predict the next text token, giving the image tokens and previous text tokens. Pretrained on COCO dataset.

How to use

Model	directory path
CNN-LSTM	Base_model/Base_model.ipynb
Blip	Blip/Blip_model.ipynb
Blip results(images with generated captions)	Blip/Blip_test.ipynb
GIT	GIT/GIT_test.ipynb
GIT results(images with generated captions)	GIT/GIT_test.ipynb

Evaluation table

Model	average Bleu
CNN-LSTM Beam-serach	0.1521
CNN-LSTM Greedy-search	0.1482
Blip	0.205
GIT	0.212

Contributors

Name	Mail
Mehrab Moradzadeh	[email protected]
Ali Derakhsesh	[email protected]
Mohammad Taha Teimuri Jervakani	[email protected]
Abolfazl malekahmadi	[email protected]
Alireza Aghaei	[email protected]