Enhancing placental pathology detection with GAMatrix-YOLOv8 model

Area of Science:

• Artificial Intelligence in Pathology
• Deep Learning for Medical Imaging
• Computer Vision in Histopathology

Background:

• Artificial intelligence (AI) techniques, including deep learning and convolutional neural networks, are increasingly applied to placental pathology.
• Current limitations exist in achieving real-time recognition and precise localization for AI-driven placental examinations.
• There is a need for enhanced AI models to improve the accuracy and efficiency of placental tissue analysis.

Purpose of the Study:

• To enhance and validate the YOLOv8 model for placental pathology.
• To investigate the significance of an improved YOLOv8 model in detecting pathological features in placental tissues.
• To develop an AI tool for real-time, accurate pathological detection in placental samples.

Main Methods:

• Integration of GAM (Gated Attention Mechanism) attention into the YOLOv8 backbone network.
• Implementation of image enhancement and normalization as preprocessing steps.
• Development of the GAMatrix-YOLOv8 model incorporating these enhancements for improved feature focus and accuracy.

Main Results:

• The GAMatrix-YOLOv8 model demonstrated superior object detection accuracy and efficiency, achieving near 100% training and validation accuracy.
• Compared to GoogleNet, ResNet18, and standard YOLOv8, GAMatrix-YOLOv8 achieved significantly higher metrics: Accuracy (0.997), Precision (0.975), Recall (0.970), and F1-Score (0.972).
• A user-friendly graphical user interface (GUI) was developed for real-time image uploading, prediction visualization, and result analysis.

Conclusions:

• GAMatrix-YOLOv8 achieves high prediction accuracy for delayed villous maturation in placental tissues via algorithmic innovation.
• The developed GUI facilitates real-time analysis and verification of pathological detection results.
• This study provides a foundation for AI-powered auxiliary diagnostic systems in pathology.