Title: RL-ACRGNet: A Reinforcement Learning Framework for Automated Chest Radiology Reporting

Abstract:

While medical imaging serves as a cornerstone of contemporary clinical diagnosis, the manual drafting of radiology reports is a labor-intensive endeavor often marred by subjective inconsistencies. In the realm of medical artificial intelligence, leveraging deep learning to automate these descriptive tasks offers the potential to optimize clinical operations and ensure standardized diagnostic outcomes. Nevertheless, achieving high-fidelity disease detection and generating clinically coherent reports remains difficult, largely due to the challenges of extracting subtle visual details and maintaining narrative consistency.

To overcome these hurdles, this study introduces RL-ACRGNet, an enhanced encoder-decoder architecture. This model combines a pre-trained DenseNet encoder with a multilevel LSTM decoder, operating within an off-policy reinforcement learning paradigm. By employing a dual-network strategy that refines visual-semantic embeddings via a metric-based reward system, we show that RL-ACRGNet surpasses current state-of-the-art benchmarks on the IU-Xray dataset. Specifically, the model yields quantitative gains of 0.47% in BLEU-4, 0.17% in METEOR, and 0.518 in ROUGE-L. Additionally, extensive testing on the large-scale MIMIC-CXR dataset validates the model’s strong generalization capabilities and its proficiency in producing high-quality, clinically pertinent reports.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC