Visual Question Answering (VQA) for Medical Imaging

Kalbe Digital Lab

Overview

The project addresses the challenge of accurate and efficient medical imaging analysis in healthcare, aiming to reduce human error and workload for radiologists. The proposed solution involves developing advanced AI models for Visual Question Answering (VQA) to assist healthcare professionals in analyzing medical images quickly and accurately. These models will be integrated into a user-friendly web application, providing a practical tool for real-world healthcare settings.

References: https://www.sciencedirect.com/science/article/abs/pii/S0933365723001252

Dataset

The model is trained with Colorectal Nuclear Segmentation and Phenotypes (CoNSeP) dataset https://warwick.ac.uk/fac/cross_fac/tia/data/hovernet. Images were extracted from 16 colorectal adenocarcinoma (CRA) WSIs.

  • Target: Nuclei
  • Task: Classification
  • Modality: Images (Histology and Label)

Model Architecture

The model is trained using DenseNet121 over CoNSep dataset.

model-architecture

Demo

Please select or upload a nuclei histology image and label image to see Nuclei Cells Classification capabilities of this model