Related Experiment Video
Updated: Jun 3, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Improving spleen segmentation in ultrasound images using a hybrid deep learning framework
Ali Karimi1, Javad Seraj1, Fatemeh Mirzadeh Sarcheshmeh2
1School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran.
Scientific Reports
|January 11, 2025
Summary
A novel two-phase method using SegFormer and Pix2Pix achieves accurate spleen segmentation in ultrasound images. This approach, validated on the Spleenex dataset, outperforms existing models, showing clinical applicability.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate spleen segmentation is crucial for diagnosing various medical conditions.
- Existing segmentation methods often struggle with the complexities and noise inherent in ultrasound images.
Purpose of the Study:
- To develop and validate a novel, highly accurate spleen segmentation method for ultrasound images.
- To introduce the Spleenex dataset, the first of its kind for spleen ultrasound segmentation.
Main Methods:
- A two-phase training approach combining SegFormerB0 for initial segmentation and Pix2Pix for refinement.
- Development and utilization of the Spleenex dataset comprising 450 spleen ultrasound images.
Main Results:
- The proposed hybrid method achieved a mean Intersection over Union (mIoU) of 94.17% and a mean Dice (mDice) score of 96.82%.
- Outperformed state-of-the-art models including SSNet, U-Net, and VAE-based methods.
- Demonstrated a Mean Percentage Length Error (MPLE) of 3.64% and robust performance on noisy images.
Conclusions:
- The novel two-phase segmentation method offers superior accuracy and robustness for spleen imaging.
- The Spleenex dataset and the proposed method advance the field of medical image analysis for spleen assessment.
