Two-Stage Liver and Tumor Segmentation Algorithm Based on Convolutional Neural Network.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Artificial Intelligence

Background:

• Malignant liver tumors pose a significant threat to human health.
• Accurate segmentation of the liver and tumors is crucial for diagnosis and treatment planning.
• Existing segmentation algorithms require improvement for precision, especially for smaller tumors.

Purpose of the Study:

• To develop and evaluate a novel two-stage convolutional neural network (CNN) algorithm for precise liver and liver tumor segmentation.
• To enhance the identification of small liver tumors using an attention mechanism.
• To validate the algorithm's performance on a public dataset.

Main Methods:

• A two-stage CNN approach was employed, starting with liver localization and followed by tumor segmentation.
• The liver localization stage utilized an encoding-decoding structure with long-distance feature fusion.
• The tumor segmentation stage incorporated 2D and 3D features, with an attention mechanism to improve small tumor detection.

Main Results:

• The algorithm achieved a Dice coefficient of 0.967 for liver segmentation.
• A Dice coefficient of 0.725 was obtained for liver tumor segmentation.
• The proposed method demonstrated superior performance compared to other state-of-the-art algorithms on the LiTS dataset.

Conclusions:

• The developed two-stage CNN algorithm accurately segments liver and liver tumors in CT images.
• The integration of an attention mechanism effectively improves the segmentation of small liver tumors.
• This algorithm represents a significant advancement in computer-aided diagnosis for liver cancer.