MPSHT: Multiple Progressive Sampling Hybrid Model Multi-Organ Segmentation.

Area of Science:

• Medical Image Analysis
• Computer-Assisted Diagnosis
• Artificial Intelligence in Healthcare

Background:

• Accurate multi-organ segmentation in Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) is crucial for disease diagnosis and treatment planning.
• Existing segmentation methods face challenges in accurately delineating multiple organs in complex medical images.

Purpose of the Study:

• To propose a novel hybrid model for enhanced multi-organ segmentation in CT and MRI images.
• To improve the accuracy and efficiency of medical image segmentation for clinical applications.

Main Methods:

• A CNN-Transformer hybrid model incorporating a progressive sampling module was developed.
• The model was evaluated on the Synapse multi-organ CT dataset and the Automated Cardiac Diagnosis Challenge (ACDC) dataset.
• Segmentation performance was assessed using Dice Similarity Coefficient (DSC) and Hausdorff_95 (HD95) metrics.

Main Results:

• The proposed model achieved an average DSC of 79.76% and HD95 of 21.55% on the Synapse CT dataset.
• Specific organ segmentation on the Synapse dataset included Kidney (R) at 80.77%, Pancreas at 59.84%, and Stomach at 81.11% DSC.
• On the ACDC MRI dataset, the model achieved an average DSC of 91.8%, outperforming existing state-of-the-art methods.

Conclusions:

• The developed multi-sampled vision transformer (MPSHT) effectively combines CNN and Transformer advantages for superior organ segmentation.
• The progressive sampling module enhances segmentation accuracy, addressing limitations of previous hybrid models.
• This approach offers a promising advancement in computer-assisted diagnosis through improved medical image segmentation.