SAA-SDM: Neural Networks Faster Learned to Segment Organ Images.

Area of Science:

• Medical Imaging Analysis
• Computer-Aided Diagnosis
• Artificial Intelligence in Medicine

Background:

• Accurate organ segmentation in medical images is vital for diagnosis and treatment planning.
• Current neural network training for segmentation can be time-consuming and resource-intensive.
• Enhancing semantic understanding and convergence speed is crucial for clinical applications.

Purpose of the Study:

• To introduce a novel feature map module, Strength Attention Area Signed Distance Map (SAA-SDM), for accelerating neural network training in medical image segmentation.
• To improve the precision and generalization performance of segmentation models.
• To reduce the computational costs associated with training deep learning models for medical imaging.

Main Methods:

• Development of the Strength Attention Area Signed Distance Map (SAA-SDM) module based on Principal Component Analysis (PCA).
• Integration of SAA-SDM to provide confidence information, enhancing semantic understanding.
• Implementation of a tailored training scheme to optimize segmentation performance.
• Validation using Transrectal Ultrasound (TRUS) and chest X-ray datasets.

Main Results:

• Significant enhancement in neural network convergence speed and precision for organ segmentation.
• Over 30% improvement in convergence speed for UNet and UNet++ models.
• Segformer achieved over 6% and 3% increase in mean Intersection over Union (mIoU) on two datasets without pre-training.
• Demonstrated effective learning guidance even without pre-trained parameters.

Conclusions:

• The SAA-SDM module effectively accelerates neural network convergence and improves segmentation accuracy.
• The proposed method reduces training time and resource requirements for medical organ segmentation.
• SAA-SDM enables high-performance segmentation without the need for pre-trained models, enhancing accessibility and efficiency.