Uterine cancer classification from CT images using convolutional feature extraction and transformer-based...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Oncology Diagnostics

Background:

• Uterine cancer diagnosis from CT images is challenging due to complex anatomy and subtle tissue variations.
• Traditional methods and standard deep learning models struggle to capture both local and global image features effectively.

Purpose of the Study:

• To develop and validate a novel hybrid deep learning framework for improved uterine cancer classification from CT images.
• To enhance the accuracy and robustness of diagnostic models by integrating convolutional and transformer-based attention mechanisms.

Main Methods:

• A hybrid framework using DenseNet121 for convolutional feature extraction and a transformer encoder with multi-head self-attention for global context.
• The model processes local features from CT scans to build enhanced global representations.
• Performance evaluated on the KAUH uterine cancer CT dataset (normal, benign, malignant) and compared against VGG16, VGG19, MobileNetV2, ResNet50, and DenseNet121.

Main Results:

• The proposed hybrid model outperformed comparative state-of-the-art deep learning models.
• Achieved high performance metrics: 87.44% accuracy, 87.13% sensitivity, 95.20% specificity, 87.17% F1 score, and 99.41% AUC.
• Demonstrated superior capability in classifying uterine tissues from CT scans.

Conclusions:

• Integrating CNNs with transformer-based self-attention significantly improves uterine cancer classification accuracy from CT images.
• The model shows potential as a computer-aided decision-support tool for radiologists in uterine cancer detection.
• The framework may be extensible to other clinical applications requiring complex image analysis.