Application of Deep Learning in Generating Structured Radiology Reports: A Transformer-Based Technique.

Since radiology reports needed for clinical practice and research are written and stored in free-text narrations, extraction of relative information for further analysis is difficult. In these circumstances, natural language processing (NLP) techniques can facilitate automatic information extraction and transformation of free-text formats to structured data. In recent years, deep learning (DL)-based models have been adapted for NLP experiments with promising results. Despite the significant potential of DL models based on artificial neural networks (ANN) and convolutional neural networks (CNN), the models face some limitations to implement in clinical practice. Transformers, another new DL architecture, have been increasingly applied to improve the process. Therefore, in this study, we propose a transformer-based fine-grained named entity recognition (NER) architecture for clinical information extraction. We collected 88 abdominopelvic sonography reports in free-text formats and annotated them based on our developed information schema. The text-to-text transfer transformer model (T5) and Scifive, a pre-trained domain-specific adaptation of the T5 model, were applied for fine-tuning to extract entities and relations and transform the input into a structured format. Our transformer-based model in this study outperformed previously applied approaches such as ANN and CNN models based on ROUGE-1, ROUGE-2, ROUGE-L, and BLEU scores of 0.816, 0.668, 0.528, and 0.743, respectively, while providing an interpretable structured report.

Association between Latest Activated Sites in the Left Ventricle and Akinetic Segments in Patients with Ischemic Cardiomyopathy.

Background: It is not clear whether the latest activation sites in the left ventricle (LV) are matched with infracted regions in patients with ischemic cardiomyopathy (ICM). We aimed to investigate whether the latest activation sites in the LV are in agreement with the region of akinesia in patients with ICM. Methods: Data were analyzed in 106 patients (age = 60.5 ± 12.1 y, male = 88.7%) with ICM (ejection fraction ≤ 35%) who were refractory to pharmacological therapy and were referred to the echocardiography department for an evaluation of the feasibility of cardiac resynchronization therapy. Wall motion abnormalities, time to peak systolic myocardial velocity (Ts) of 6 basal and 6 mid-portion segments of the LV, and 4 frequently used dyssynchrony indices were measured using 2-dimensional echocardiography and tissue Doppler imaging (TDI). To evaluate the influence of the electrocardiographic pattern, we categorized the patients into 2 groups: patients with QRS ≤ 120 ms and those with QRS >120 ms. Results: A total of 1 272 segments were studied. The latest activation sites (with longest Ts) were most frequently located in the mid-anterior (n = 32, 30.2%) and basal-anterior segments (n = 29, 27.4%), while the most common sites of akinesia were the mid-anteroseptal (n = 65, 61.3%) and mid-septal (n = 51, 48.1%) segments. Generally, no significant concordance was found between the latest activated segments and akinesia either in all the patients or in the QRS groups. Detailed analysis within the segments indicated a good agreement between akinesia and delayed activation in the basal-lateral segment solely in the patients with QRS duration ≤ 120 ms (Φ = 0.707; p value ≤ 0.001). Conclusion: The akinetic segment on 2-dimensional echocardiogram was not matched with the latest activation sites in the LV determined by TDI in patients with ICM.