Fatty-Liver-Level-Recognition-Using-Particle-Swarm-optimization-PSO-Image-Segmentation-and-Analysi

%% Fatty Liver Level Recognition Using Particle Swarm optimization (PSO) Image Segmentation and Analysis

• Link to the paper:
• https://ieeexplore.ieee.org/document/9960108
• DOI: 10.1109/ICCKE57176.2022.9960108

Please cite below:

Mousavi, Seyed Muhammad Hossein, et al. "Fatty liver level recognition using particle swarm optimization (PSO) image segmentation and analysis." 2022 12th International Conference on Computer and Knowledge Engineering (ICCKE). IEEE, 2022.

Fatty Liver Level Recognition Using Particle Swarm Optimization (PSO) Image Segmentation

This repository contains the implementation and details of the Fatty Liver Level Recognition system using Particle Swarm Optimization (PSO) and other image segmentation techniques. The proposed system efficiently segments microscopic liver images to detect fat deposits and classify fatty liver levels.

---

Introduction

Fatty liver disease, caused by excessive fat deposits in the liver, is a significant health issue. This system leverages Particle Swarm Optimization (PSO) and various image segmentation techniques to detect fatty liver and classify its severity into different levels. The implementation focuses on high-resolution microscopic images with a zoom level of 200x or more.

---

Features

• Segmentation Methods:
  • Particle Swarm Optimization (PSO)
  • Otsu's Thresholding
  • Watershed Algorithm
  • K-Means Clustering
• Performance Metrics:
  • Accuracy
  • F-Score
  • Intersection over Union (IoU)
• Visualization:
  • Segmented liver images
  • Fat deposit recognition and classification levels.

---

Workflow

1. Preprocessing:
   • Intensity adjustment
   • Histogram equalization
   • Canny edge detection
2. Segmentation:
   • Segment images using PSO, Otsu, Watershed, and K-Means.
3. Fatty Liver Level Recognition:
   • Analyze segmented images to detect fat deposits and classify fatty liver levels based on predefined markers.

---

Performance

The system achieved remarkable results when compared with traditional segmentation methods:

• Average Accuracy: 92.2%
• Average F-Score: 87.2%
• Average IoU: 90.7%

The PSO algorithm consistently outperformed Otsu, Watershed, and K-Means in all metrics.

---

Segmentation Techniques

1. Otsu's Thresholding: Minimizes interclass variance for binary segmentation.
2. Watershed Algorithm: Region-based segmentation inspired by drainage patterns.
3. K-Means Clustering: Clusters image pixels based on intensity or color.
4. Particle Swarm Optimization (PSO): Nature-inspired optimization algorithm, providing superior segmentation results.

---

• Link to the paper:
• https://ieeexplore.ieee.org/document/9960108
• DOI: 10.1109/ICCKE57176.2022.9960108