Evolutionary-based deep learning network model using adaptive mixing differential evolution and application in acute...

Area of Science:

• Artificial Intelligence in Medicine
• Computational Biology
• Medical Informatics

Background:

• Acute pulmonary embolism (APE) presents with non-specific symptoms, leading to diagnostic challenges and high mortality.
• Current risk stratification methods for APE are often complex, invasive, and lack repeatability.
• There is a critical need for efficient and accurate tools for APE prediction and analysis.

Purpose of the Study:

• To develop an evolutionary-based deep learning network, EDLAlexNet, for precise prediction and analysis of APE patients.
• To utilize accessible clinical data, including blood biochemical indices, vital signs, and clinical characteristics.
• To provide a reliable clinical tool for APE assessment with high accuracy, specificity, sensitivity, and AUC.

Main Methods:

• Developed the EDLAlexNet model integrating an adaptive mixing differential evolution (MIXDE) evolutionary computation method.
• Incorporated Q-learning and opposition-based learning within the MIXDE algorithm.
• Validated the MIXDE algorithm's performance on standard datasets and applied the EDLAlexNet model to analyze patient data.

Main Results:

• EDLAlexNet achieved high performance in APE prediction: 93.76% accuracy, 89.46% specificity, and 95.74% sensitivity.
• The model demonstrated a significant area under the curve (AUC) of 0.9527.
• These results confirm the model's effectiveness in precisely predicting and analyzing APE patients.

Conclusions:

• The EDLAlexNet model, incorporating the MIXDE algorithm, shows excellent performance in APE prediction and analysis.
• This model addresses limitations of current APE assessment methods.
• EDLAlexNet holds potential as a valuable clinical tool for APE assessment and management.