APESTNet with Mask R-CNN for Liver Tumor Segmentation and Classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate liver tumor segmentation and classification are crucial for diagnosing and treating hepatocellular carcinoma and metastases.
• Challenges in liver tumor analysis include indistinct borders and variations in shape, size, and position.
• Advancements in artificial intelligence, particularly transformer models, offer new possibilities for computer vision tasks.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for automated liver tumor segmentation and classification.
• To improve the precision of liver tumor analysis beyond existing clinical methods.
• To apply a three-stage deep learning approach for enhanced diagnostic capabilities.

Main Methods:

• A three-stage process involving pre-processing, liver segmentation, and classification was employed.
• Computed Tomography (CT) images underwent contrast enhancement (histogram equalization) and noise reduction (median filter).
• An enhanced Mask R-CNN model performed liver segmentation, followed by an Enhanced Swin Transformer Network with Adversarial Propagation (APESTNet) for classification and overfitting prevention.

Main Results:

• The proposed APESTNet model demonstrated superior performance in segmenting and classifying liver tumors across a variety of CT images.
• The model proved to be efficient and exhibited low sensitivity to noise in the CT scans.
• Experimental results validated the effectiveness of the deep learning approach for liver tumor analysis.

Conclusions:

• The novel deep learning model, APESTNet, offers a highly effective and efficient solution for liver tumor segmentation and classification.
• The proposed method enhances diagnostic accuracy for hepatocellular carcinoma and metastases.
• The model's robustness to noise and performance on diverse CT images highlight its clinical potential.