Intelligent system based on multiple networks for accurate ovarian tumor semantic segmentation.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Deep learning for oncology

Background:

• Ovarian tumors, particularly malignant types, are a growing global health concern.
• Accurate and early diagnosis is critical for effective treatment and improved survival rates.
• Existing diagnostic systems require enhancement for greater efficiency and precision.

Purpose of the Study:

• To develop a more accurate diagnostic system for ovarian tumors by combining convolutional neural networks (CNNs).
• To investigate custom combination and selection approaches for ensemble deep learning models.
• To evaluate whether combining all networks or only the best-performing ones yields superior results.

Main Methods:

• Utilized five DeepLab-V3+ networks with varied encoders (ResNet-18, ResNet-50, MobileNet-V2, InceptionResNet-V2, Xception).
• Developed a custom algorithm for combining multiple semantic segmentation networks.
• Implemented an iterative selection approach to optimize the ensemble model composition.

Main Results:

• The proposed ensemble system achieved a 91.18% Intersection over Union (IoU) for ovarian tumor semantic segmentation.
• The combined network approach outperformed all individual networks used in the study.
• The system demonstrated effectiveness for both benign and malignant ovarian tumor types.

Conclusions:

• Ensemble deep learning models, through custom combination and selection, can significantly enhance ovarian tumor diagnostic accuracy.
• The developed system offers a promising advancement for medical support in ovarian tumor detection.
• Future work can explore integrating more powerful deep learning models for further performance improvements.