- Research Article
- 10.1080/02533839.2025.2558180
An approach to medical visual question generation based on question categories
- Oct 16, 2025
- Journal of the Chinese Institute of Engineers
- N Sheerin Sitara + 3 more +3
The Visual Question Generation (VQG) is a research field that aims to generate natural language questions for given images. These images are of two types, namely natural and medical images. VQG for medical images has gained attention in the last few years for two reasons: (i) It allows the user to ask the question based on the input image, (ii) It reduces the burden on medical annotators. Even though there has been considerable improvement in Medical VQG (MVQG), it is still in the developing stage. This is because the dataset scarcity problem and a generalized model for different VQG questions are yet to be resolved. To address these issues, a MVQG system is proposed to reduce the data scarcity problem to some extent. The developed MVQG system generates all possible questions based on different categories (organ, plane, modality and abnormality) using deep learning techniques. The Convolutional Neural Network (CNN), Multi-Layer Perceptron (MLP) and Long Short Term Memory (LSTM) along with latent space encoding are used in generating questions for three datasets namely, VQA-RAD, VQA-MED 2019 and VQA-MED 2021. As a result, the proposed approach attained the BLEU score of 37.0%, 35.5% and 31.6% for VQA-RAD, VQA-MED 2019 and VQA-MED 2021 datasets respectively. The questions generated from the proposed MVQG approach can be used to develop an improved Medical Visual Question Answering (MVQA) system.
Read more