CSAMDT: Conditional Self Attention Memory-Driven Transformers for Radiology Report Generation from Chest X-Ray

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Imaging Analysis
• Radiology Informatics

Background:

• Radiology report generation is time-consuming and requires specialized expertise.
• Current AI research often focuses on image captioning, neglecting specific report elements.
• There is a need for automated systems to assist radiologists and improve efficiency.

Purpose of the Study:

• To introduce a novel Conditional Self Attention Memory-Driven Transformer model for automated radiological report generation.
• To evaluate the model's performance against existing state-of-the-art techniques.
• To demonstrate the potential of AI in alleviating radiologist workload and enhancing clinical workflows.

Main Methods:

• A two-phase approach was employed: multi-label classification with ResNet152 v2 for feature extraction and disease diagnosis, followed by a Conditional Self Attention Memory-Driven Transformer as a decoder.
• The model utilizes self-attention memory-driven transformers for text generation.
• Performance was evaluated using Bilingual Evaluation Understudy (BLEU) scores (BLEU 1-4).

Main Results:

• The proposed Conditional Self Attention Memory-Driven Transformer model achieved superior performance compared to existing methods.
• The model demonstrated improved BLEU scores: BLEU 1 (0.475), BLEU 2 (0.358), BLEU 3 (0.229), and BLEU 4 (0.165).
• These results indicate a significant advancement in automated radiological report generation.

Conclusions:

• The developed AI model offers an effective solution for autonomous radiological report generation.
• This technology has the potential to significantly reduce the burden on radiologists.
• Implementation of this system can lead to enhanced efficiency and improved clinical workflows in radiology departments.