An efficient modular framework for automatic LIONC classification of MedIMG using unified medical language

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Biomedical Data Analysis

Background:

• Clinical text data, including handwritten prescriptions and radiological reports, requires specific labeling for effective use.
• Semantic annotation of biomedical texts is crucial for advancing medical research and clinical applications.
• Deep Learning (DL) methods show promise for expert-level accuracy in automated medical image analysis.

Purpose of the Study:

• To develop a methodology for learning effective holistic representations from medical images (MedIMG), such as handwritten prescriptions and radiology reports.
• To investigate information extraction from narrative MedIMG and automatic categorization based on image resolution.
• To propose a hybrid model for Named Entity Recognition (NER) predictions using Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Module (GRM).

Main Methods:

• Utilized Deep Learning (DL) techniques for automated analysis of medical images and associated clinical text.
• Employed downscaled input images and examined functional, responsive areas and class activation maps for model training.
• Developed a hybrid model combining RNN, LSTM, and GRM for Named Entity Recognition (NER) and employed attention mechanisms for supervised training.

Main Results:

• The proposed methodology effectively extracts information from narrative MedIMG and performs automatic categorization.
• The hybrid RNN + LSTM + GRM model demonstrated strong performance in Named Entity Recognition for various input purposes.
• MetaMapLite achieved comparable recall, precision, and F1-scores for biomedical text search and medical text examination.

Conclusions:

• Deep Learning techniques are effective for generating large-scale labeled clinical data and improving medical data quality.
• The developed methods contribute to the efficient extraction and categorization of information from medical images and reports.
• This research highlights the significance of real-time efforts in biomedical studies for global appeal and diffusion.