GHS-NET a generic hybridized shallow neural network for multi-label biomedical text classification.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• The exponential growth of biomedical literature and clinical data necessitates advanced computational methods for insight extraction and disease coding.
• Current deep learning models for biomedical text classification often rely on either convolutional neural networks (CNNs) for feature extraction or recurrent neural networks (RNNs) for contextual information, but not both.
• Existing methods show limitations in classifying diverse biomedical text genres, such as scientific literature and clinical notes.

Purpose of the Study:

• To introduce GHS-NET, a novel hybrid deep learning model for generic, multi-label classification of biomedical text across different genres.
• To leverage the strengths of both CNNs and RNNs to enhance the accuracy and robustness of biomedical text classification.
• To evaluate the effectiveness of GHS-NET on extreme multi-label literature classification and clinical note coding tasks.

Main Methods:

• GHS-NET employs a hybrid architecture combining a convolutional neural network (CNN) for discriminative feature extraction and a bi-directional Long Short-Term Memory (BiLSTM) network for capturing contextual information.
• The model was evaluated on two distinct tasks: extreme multi-label classification of biomedical literature (Hallmarks of Cancer, Chemical Exposure datasets) and ICD-9 code assignment for clinical notes (MIMIC-III dataset).

Main Results:

• For extreme multi-label literature classification, GHS-NET demonstrated performance improvements over state-of-the-art methods, with notable gains in precision, recall, and F1-score on the Hallmarks of Cancer and Chemical Exposure datasets.
• In clinical note classification using the MIMIC-III dataset, GHS-NET significantly outperformed previous deep learning approaches, achieving substantial improvements in recall and F1-score.
• The proposed GHS-NET model shows significant potential for accurately classifying multi-variate disease and chemical exposure-specific text.

Conclusions:

• GHS-NET represents a significant advancement in biomedical text classification by effectively integrating CNNs and BiLSTMs.
• The model's ability to handle diverse biomedical text genres and its superior performance highlight its utility in enhancing biomedical and bioinformatics applications.
• GHS-NET offers a versatile and accurate solution for extracting insights from the rapidly expanding volume of biomedical data.