Accessing Artificial Intelligence for Fetus Health Status Using Hybrid Deep Learning Algorithm (AlexNet-SVM) on...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Cardiotocographic (CTG) recordings are crucial for fetal well-being assessment.
• Existing AI models for CTG analysis often overlook computational time, a critical factor in clinical settings.
• Accurate and timely interpretation of CTG data is essential for informed patient outcomes.

Purpose of the Study:

• To develop a modified deep neural algorithm for classifying pathological and suspicious CTG recordings with reduced time complexity.
• To enhance the efficiency of AI models in digital health applications.
• To address the limitations of previous studies by considering both accuracy and computational time.

Main Methods:

• A hybrid deep learning model merging AlexNet architecture with Support Vector Machines (SVMs) at fully connected layers.
• Utilizing a deep transfer learning (TL) mechanism with partially trained convolutional layers.
• Training on an open-source UCI dataset of 2126 CTG recordings with 23 attributes, categorized into Normal, Pathological, and Suspected classes.
• Employing an ADAM optimizer for hyperparameter optimization.

Main Results:

• The proposed algorithm achieved high real-time performance metrics: 99.72% accuracy, 96.67% sensitivity, and 99.6% specificity.
• Demonstrated superior performance compared to established architectures like RCNNs, ResNet, DenseNet, and GoogleNet.
• Significantly reduced computational time compared to traditional methods.

Conclusions:

• The developed AI algorithm offers a viable solution for real-time CTG classification in clinical settings.
• The hybrid AlexNet-SVM approach effectively balances accuracy and computational efficiency.
• Further real-time validation is recommended to confirm its clinical utility in improving patient outcomes.