FTT: Fourier Transform Based Transformer for Brain CT Report Generation

The interpretation of Brain Computed Tomography (CT) scans predominantly falls under the purview of specialized radiologists. However, given the challenges associated with excessive workloads, human resource limitations, urgency in emergency scenarios, and inconsistencies in outsourced inter-pretations, the margin for diagnostic errors is substantial. To ameliorate this issue, burgeoning research has been directed towards the automatic synthesis of various medical diagnostic reports. Contrary to conventional image captioning tasks, the domain of medical report generation is fraught with inherent biases, making it arduous to accurately extract features pertinent to specific pathological lesions. Moreover, redundant descriptions of normative areas further impede the precise delineation of anomalies. To address these challenges, this paper introduces a novel transformer architecture that synergizes lesion detection algorithms with Fourier Transform techniques. Experimental results indicate that our proposed model outperforms existing combined-embedding models and exhibits enhanced performance when applied to Fourier-transformed image data.