Multiple perception contrastive learning for automated ovarian tumor classification in CT images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Ovarian cancer is a leading cause of cancer death in women globally.
• Early detection significantly improves patient survival rates.
• Current CT image analysis relies on subjective radiologist interpretation, leading to variability.

Purpose of the Study:

• To develop an automated and reliable diagnostic system for ovarian tumour classification using CT images.
• To enhance the accuracy and robustness of ovarian cancer diagnosis through deep learning.

Main Methods:

• Utilized supervised contrastive learning and a Multiple Perception Encoder (MP Encoder) for automated CT image classification.
• Incorporated T-Pro technology for data augmentation and semantic perturbations for improved generalization.
• Integrated Multi-Scale Perception (MSP Module) and Multi-Attention (MA Module) to enhance sensitivity to tumor morphology.

Main Results:

• Achieved an average classification accuracy of 98.43% for ovarian tumours.
• Demonstrated exceptional efficacy in classifying complex or low-quality CT images.
• The deep learning framework significantly improved classification accuracy and robustness.

Conclusions:

• The developed deep learning framework offers a reliable automated diagnostic tool for ovarian tumours.
• This approach enhances diagnostic support for clinicians, aiding in early detection and treatment optimization.
• The method shows promise in overcoming challenges in interpreting complex ovarian tumour CT images.