Hybrid architecture based intelligent diagnosis assistant for GP.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• General Practitioners (GPs) are the primary point of contact in healthcare systems like the NHS.
• Accurate primary diagnosis by GPs is essential for efficient patient management and reducing specialist workload.
• GPs' broad yet less specialized knowledge can limit diagnostic accuracy, necessitating intelligent assistance.

Purpose of the Study:

• To develop an AI-powered system to assist General Practitioners (GPs) in accurate primary diagnosis.
• To improve the accuracy and efficiency of diagnosing common neurological diseases in primary care settings.
• To enhance the integration of patient symptom data for improved diagnostic performance.

Main Methods:

• Introduction of two data augmentation techniques: Complaint Symptoms Integration Method and Symptom Dot Separating Method.
• Proposal of a hybrid deep learning architecture fusing features from different word representation spaces.
• Development of an AI diagnosis assistant web application utilizing the proposed hybrid architecture.

Main Results:

• The enhanced Integration dataset, incorporating data augmentation, was used for training and evaluation.
• The hybrid architecture demonstrated superior classification performance for four common neurological diseases compared to standard models.
• The BERT+CNN hybrid architecture achieved 89.7% classification accuracy, a 5.1% improvement over BERT and CNN alone.

Conclusions:

• The developed AI diagnosis assistant effectively aids GPs in primary diagnosis, improving efficiency and accuracy.
• The novel data augmentation methods and hybrid architecture significantly enhance diagnostic performance for neurological conditions.
• This AI system has the potential to optimize patient pathways and reduce the burden on specialist services.