Enhanced boundary-directed lightweight approach for digital pathological image analysis in critical oncological...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Pathological image analysis is vital for diagnosing critically ill cancer patients, but time constraints and image complexity challenge timely interventions.
• Current deep learning models in medical decision support systems face computational challenges, hindering real-time emergency diagnostics.
• Accurate and rapid diagnosis is essential to avoid missing the optimal treatment window for cancer patients.

Purpose of the Study:

• To develop an efficient and accurate diagnostic strategy for malignant bone tumors in emergency settings.
• To address the limitations of current deep learning models in terms of computational complexity and real-time processing for pathological image analysis.

Main Methods:

• Proposed a Lightened Boundary-enhanced Digital Pathological Image Recognition Strategy (LB-DPRS) for osteosarcoma diagnosis.
• Optimized Transformer's self-attention mechanism with boundary segmentation enhancement for improved tissue and nuclear boundary recognition.
• Implemented row-column attention and complementary attention mechanisms to reduce computational load and enhance feature extraction.

Main Results:

• The LB-DPRS strategy achieved a Dice Similarity Coefficient (DSC) of 0.862 and an Intersection over Union (IOU) of 0.749.
• The model demonstrated high computational efficiency with only 10.97 million parameters.
• Significant improvements in computational efficiency and prediction accuracy were observed.

Conclusions:

• The LB-DPRS strategy offers a powerful and efficient solution for the emergency diagnosis of malignant bone tumors like osteosarcoma.
• The approach enhances model interpretability while maintaining high diagnostic performance.
• This strategy supports timely and accurate clinical decision-making in critical care settings.