A Lightweight Convolutional Neural Network Model for Liver Segmentation in Medical Diagnosis.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Liver segmentation from CT images is crucial for medical diagnosis and treatment planning.
• Current deep learning methods for liver segmentation are computationally intensive and require substantial hardware resources, limiting their accessibility.
• There is a need for efficient and lightweight models for liver segmentation in medical imaging.

Purpose of the Study:

• To propose a lightweight convolutional neural network (CNN) for accurate and efficient liver segmentation from CT images.
• To address the computational challenges and hardware limitations associated with existing deep learning models for this task.
• To introduce a novel network architecture, Ga-CNN, utilizing Gaussian weight initialization.

Main Methods:

• Developed a lightweight CNN architecture (Ga-CNN) with 3 convolutional and 2 fully connected layers.
• Employed softmax for liver and background discrimination.
• Utilized random Gaussian distribution for weight initialization to achieve distance-preserving embeddings.

Main Results:

• The proposed Ga-CNN model demonstrated effective liver segmentation performance across three benchmark datasets: MICCAI SLiver'07, 3Dircadb01, and LiTS17.
• Experimental results indicate that the lightweight nature of Ga-CNN does not compromise its segmentation accuracy.
• The method achieved good performance, suggesting its viability for clinical applications.

Conclusions:

• The proposed Ga-CNN is an efficient and lightweight deep learning model for liver segmentation in CT images.
• Ga-CNN offers a practical solution for medical practitioners by reducing training time and hardware requirements.
• This approach advances the field of medical image analysis by providing an accessible tool for liver segmentation.