Semantic Segmentation of Urinary Bladder Cancer Masses from CT Images: A Transfer Learning Approach.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Urinary bladder cancer is a prevalent malignancy with high metastatic and recurrence rates.
• Accurate and timely diagnosis is critical for effective treatment and patient outcomes.
• Artificial intelligence (AI) offers potential for enhancing diagnostic accuracy in clinical settings.

Purpose of the Study:

• To develop and evaluate an AI-based semantic segmentation system for urinary bladder cancer masses from CT images.
• To assess the performance of transfer learning approaches using various deep learning architectures for tumor segmentation.
• To investigate the system's efficacy across different image planes (frontal, axial, sagittal).

Main Methods:

• A transfer learning approach was applied to semantic segmentation of urinary bladder cancer from CT scans.
• The dataset was divided into frontal, axial, and sagittal image planes.
• AlexNet was used for plane recognition, and U-net with pre-trained backbones (ResNet101, ResNet50, VGG-16) was employed for segmentation.

Main Results:

• The plane recognition system achieved high performance (AUCmicro = 0.9999).
• Semantic segmentation using U-net with pre-trained backbones demonstrated high performance across all planes (DSC up to 0.9660).
• ResNet101, ResNet50, and VGG-16 backbones yielded optimal results for frontal, axial, and sagittal planes, respectively.

Conclusions:

• The proposed AI system for semantic segmentation of urinary bladder cancer masses from CT images exhibits high performance and generalization capabilities.
• The findings suggest the potential clinical utility of this AI system for improving bladder cancer diagnosis.
• This AI-driven approach could significantly contribute to more accurate and timely cancer detection in clinical practice.