Title: An efficient medical emergency prediction model from unstructured medical transcript using dual attentional Bi-LSTM model

Authors: Amita Mishra; Sunita Soni

Addresses: Department of Computer Science and Engineering, CSVTU, India; Department of Computer Science and Engineering, Anurag University, India ' Department of Computer Science and Engineering, BIT-DURG, India

Abstract: This study tackles the challenge of predicting medical emergencies from unstructured transcripts, enhancing clinical decision-making. Employing a dual attention Bi-LSTM model, it integrates pre-processing, feature extraction, and summarisation steps. A weighted hybrid distance-based graph embedding technique captures relevant features, and a fine-tuned BERT model with dual attention effectively summarises the text. Additionally, an adaptive rule-based Bi-LSTM captures temporal connections and contextual information. The model performs exceptionally well with 96.44% accuracy, 98.18% sensitivity, 95.95% specificity, and a 96.42% F1-score for classification. Summarisation results surpass existing methods with BLEU 0.43, CIDER 0.74, METEOR 0.22, ROUGE 0.53, and SPICE 19.74. This dual attention Bi-LSTM model holds promise for enhancing clinical workflows by extracting critical information from medical transcripts.

Keywords: deep Bi-LSTM; dual attention Bi-LSTM; BERT; TF-IDF; weighted hybrid distance-based graph embedding.

DOI: 10.1504/IJBRA.2025.150105

International Journal of Bioinformatics Research and Applications, 2025 Vol.21 No.6, pp.623 - 639

Received: 17 Apr 2024
Accepted: 01 Aug 2024
Published online: 01 Dec 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article