Performance Analysis of Segmentation and Classification of CT-Scanned Ovarian Tumours Using U-Net and Deep...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early detection of ovarian cancer remains a significant challenge, contributing to high mortality rates despite treatment advancements.
• Current diagnostic methods for ovarian tumors can be limited in their ability to detect early-stage malignancies.

Purpose of the Study:

• To evaluate the efficacy of deep learning algorithms for early ovarian tumor detection and classification using CT scan images.
• To compare the performance of various convolutional neural network (CNN) architectures and traditional machine learning models in differentiating benign from malignant ovarian tumors.

Main Methods:

• CT scan images of the ovarian region were pre-processed and tumors were segmented using the UNet model.
• Deep learning models (CNN, ResNet, DenseNet, Inception-ResNet, VGG16, Xception) and machine learning models (Random Forest, Gradient Boosting, AdaBoosting, XGBoosting) were employed for classification.
• Optimization techniques were applied to machine learning models to enhance performance.

Main Results:

• DenseNet 121 demonstrated superior performance, achieving an accuracy of 95.7% in classifying ovarian tumors.
• The study provided a comparative analysis of multiple CNN architectures against common machine learning algorithms, with and without optimization.

Conclusions:

• Deep learning models, particularly DenseNet 121, show significant potential as effective diagnostic tools for early ovarian cancer detection from CT images.
• The findings highlight the advantage of using advanced AI techniques for improved accuracy in classifying benign and malignant ovarian tumors.