Liver Tumor Localization Based on YOLOv3 and 3D-Semantic Segmentation Using Deep Neural Networks

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Liver cancer causes over 1.5 million deaths globally each year.
• Early detection via computed tomography (CT) is crucial for improving patient outcomes.
• Accurate segmentation of small liver tumors in CT scans is challenging due to variations in size, shape, intensity, and contrast.

Purpose of the Study:

• To develop a robust, automated system for detecting and segmenting minute liver tumors in CT images.
• To enhance the reliability and accuracy of computer-aided diagnosis for liver cancer.

Main Methods:

• A three-part model was developed: synthetic image generation using a generative adversarial network (GAN), tumor localization using a Resnet-50 and YOLOv3-based approach, and segmentation using a DeepLabv3 model with an InceptionResNetv2 backbone.
• The localization model achieved 0.99 mean average precision (mAP).
• The segmentation model achieved over 95% accuracy in the testing phase.

Main Results:

• The proposed AI model demonstrated high accuracy in localizing and segmenting liver tumors, including minute ones.
• The system outperformed recently published methods in terms of accuracy for liver and tumor localization and segmentation.
• Achieved >95% accuracy in the testing phase.

Conclusions:

• The developed AI model offers a promising solution for accurate and reliable automated analysis of liver CT scans.
• This approach has the potential to significantly aid in the early detection and diagnosis of liver cancer.
• The study highlights the effectiveness of combining GANs, deep feature extraction, and advanced segmentation networks for medical image analysis.