Ovarian cancer identification technology based on deep learning and second harmonic generation imaging.

Area of Science:

• Biomedical Imaging
• Oncology
• Artificial Intelligence

Background:

• Ovarian cancer is a leading cause of cancer death in women globally.
• Current biopsy methods for ovarian cancer screening are time-consuming and require expert interpretation.
• There is a need for rapid, accurate, and objective diagnostic tools for ovarian cancer.

Purpose of the Study:

• To develop a simple, fast, and label-free method for ovarian cancer diagnosis.
• To evaluate the efficacy of second harmonic generation (SHG) imaging combined with deep learning for characterizing ovarian tissues.
• To accurately differentiate benign, normal, and malignant ovarian tissues using SHG imaging and AI.

Main Methods:

• Utilized second harmonic generation (SHG) imaging to capture unstained fresh human ovarian tissues.
• Employed the Pyramid Vision Transformer V2 (PVTv2) deep learning model for image analysis and classification.
• Quantified ovarian cancer by analyzing collagen fiber structures within SHG images.

Main Results:

• SHG imaging effectively visualized collagen fibers, enabling quantification of ovarian cancer.
• The PVTv2 model achieved a 98.4% accuracy in classifying 3240 SHG images into benign, normal, and malignant categories.
• Demonstrated the potential of SHG imaging and deep learning for objective ovarian tissue diagnosis.

Conclusions:

• Second harmonic generation (SHG) imaging provides a label-free approach to visualize tissue morphology relevant to ovarian cancer.
• Deep learning models, specifically PVTv2, can accurately interpret SHG images for rapid and objective ovarian cancer diagnosis.
• This combined approach shows significant promise for improving the efficiency and accuracy of ovarian cancer screening during surgery.