A cross-modal Chinese radiology report generation approach for cervical cancer

Abstract

Magnetic resonance imaging (MRI) is widely used in the pathological evaluation and early diagnosis of cervical cancer (CC). Conventional automatic report generation approaches are predominantly designed for single-image analysis, limiting their applicability to MRI sequences that inherently contain richer temporal and spatial information. Furthermore, sequence-based features may introduce redundancy and noise, challenging model robustness. In this study, we propose a CC Chinese report generation method (C3RG) tailored for CC MRI sequences. The proposed framework incorporates a feature refinement network (FRN) to suppress redundant channel information and enhance salient feature representation. In addition, a cross-modal memory network (CMN) and an interactive feed-forward network (IFFN) are integrated into both the encoder and decoder to facilitate efficient multimodal interaction and alignment between image and text modalities. The model is built upon a Transformer-based encoder–decoder architecture. To support training and evaluation, we construct a dedicated dataset consisting of CC MRI sequences and their corresponding Chinese diagnostic reports. Experimental results demonstrate that C3RG outperforms existing state-of-the-art models, achieving BLEU-1, BLEU-2, BLEU-3, BLEU-4, ROUGE-L, and CIDEr scores of 0.458, 0.319, 0.226, 0.165, 0.379, and 0.264, respectively. Ablation studies further confirm the contribution of each component. These results indicate that C3RG holds promise for clinical deployment in automated radiology reporting for CC.