A lightweight transformer-based hybrid encoder-decoder model for chest X-ray medical report generation.

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Natural Language Generation

Background:

• Automated medical report generation from images is challenging for deep learning.
• Existing methods often neglect computational efficiency for high accuracy.
• Accurate and efficient systems are needed to support clinical decision-making.

Purpose of the Study:

• To develop a low computational cost, high-performance hybrid encoder-decoder architecture for autonomous medical report generation.
• To create a system capable of producing paragraph-level reports from medical images.
• To enhance diagnostic support for physicians.

Main Methods:

• A hybrid encoder-decoder architecture named FAST-MRG was developed.
• The encoder uses a transformer with distillation for feature extraction.
• The decoder employs a generative pre-trained transformer for text generation.
• Trained and tested on the Indiana University Chest X-ray dataset.

Main Results:

• FAST-MRG achieved scores of 0.373 (Bleu-1), 0.226 (Meteor), and 0.332 (Rouge).
• Demonstrated an average time efficiency of 66% compared to prior work.
• Produced meaningful reports supporting diagnosis and treatment.

Conclusions:

• The FAST-MRG architecture offers a balance of high performance and low computational cost.
• It effectively generates paragraph-level medical reports from chest X-rays.
• The model shows promise as a foundation for future research in autonomous medical reporting.