TPC6323-Project

📋 Project Overview

Subject: TPC6323 Parallel Computing

Title: Parallel Image Segmentation using K-Means for Brain Tumour

Category: Signal Processing

This project investigates the use of parallel computing techniques for brain tumour MRI image segmentation. Five segmentation algorithms are implemented and evaluated under three different execution models:

• Sequential CPU
• CPU Parallelism (Parallel.For)
• GPU Parallelism (ILGPU)

The objective is to analyse computational performance improvements achieved through parallel computing while preserving the original segmentation algorithms. Execution time, speedup, scalability, and efficiency are compared to determine the most suitable execution model for medical image processing applications.

---

🎯 Objectives

• Implement five brain tumour segmentation algorithms.
• Develop sequential, CPU-parallel, and GPU-parallel versions of each algorithm.
• Measure execution time using BenchmarkDotNet.
• Compare computational speedup between execution models.
• Demonstrate the benefits of parallel computing in medical image segmentation.

---

🧠 Algorithms Implemented

Algorithm	Description
K-Means	Clustering algorithm based on pixel intensity similarity.
Mean Shift	Density-based clustering for image segmentation.
Fuzzy C-Means	Soft clustering where pixels belong to multiple clusters with different membership values.
SLIC Superpixels	Generates superpixels to simplify image segmentation.
Graph-Based Segmentation	Segments images by modelling pixels as graph nodes and minimizing edge weights.

---

⚡ Execution Models

Execution Model	Description
Sequential CPU	Traditional single-threaded implementation.
CPU Parallel	Multi-threaded implementation using Parallel.For.
GPU Parallel	GPU implementation using ILGPU for massive parallel execution.

---

✨ Key Features

• Five Segmentation Algorithms: K-Means, Mean Shift, Fuzzy C-Means, SLIC, and Graph-Based
• Three Execution Models: Sequential CPU, Multi-threaded CPU (Parallel.For), GPU Parallel (ILGPU)
• Comprehensive Performance Benchmarking: Using BenchmarkDotNet for accurate measurements
• End-to-End Pipeline: From image preprocessing to segmentation mask generation
• Dataset: Kaggle Brain Tumor MRI Dataset with glioma, pituitary, and meningioma classes

---

🛠 Technology Stack

Category	Technology
Programming Language	C#
Framework	.NET
GPU Computing	ILGPU
Parallel Programming	Task Parallel Library (TPL)
Benchmarking	BenchmarkDotNet
Image Processing	System.Drawing

---

📊 Dataset

The project uses publicly available Brain MRI datasets from Kaggle.

📂 Dataset Sources

• Brain Tumor MRI Dataset

  • https://www.kaggle.com/datasets/masoudnickparvar/brain-tumor-mri-dataset

• Grad-CAM Segmentation Dataset

  • https://www.kaggle.com/code/noorsaeed/99-mri-classification-with-grad-cam-segmentation

🧠 Tumour Classes

Class	Description
Glioma	Malignant brain tumour
Meningioma	Tumour arising from the meninges
Pituitary	Tumour located in the pituitary gland

---

📁 Project Structure

TPC6323-Project/
│
├── SLIC/                                       # SLIC Superpixels segmentation (Algorithm)
│   ├── Benchmark-focusOnAlgOnly/               # BenchmarkDotNet performance evaluation (CPU, GPU, Sequential & Prototype)
│   ├── Dataset/                                # Brain MRI datasetssegmentation (glioma, meningioma & pituitary)
│   ├── Grad-CAM-based pseudo ground truth/     # Pseudo ground truth generated using Grad-CAM (glioma, meningioma & pituitary)
│   ├── Parallel Slic (Full Benchmark Check)/   # Parallel SLIC (CPU & GPU)
│   └── TPC Preprocessing/                      # Image preprocessing utilities (Dataset, Preprocess Dataset & Preprocessing Code)
│
├── Fuzzy_C_Means/                              # Fuzzy C-Means (Algorithm)
│   ├── TPC_Fuzzy_CPU/                          # CPU Parallel
│   └── TPC_Fuzzy_GPU/                          # GPU Parallel 
│
├── GBS/                                        # Graph-Based Segmentation (Algorithm)
│   ├── backup/                                 # Backup source codes
│   ├── GBS_CPU/                                # CPU Parallel
│   ├── GBS_GPU/                                # GPU Parallel
│   └── Sequential GBS/                         # Sequential (old)
│
├── K-Means/                                    # K-Means segmentation (Algorithm)
│   ├── Baseline K-Means                        # Sequential CPU
│   ├── CPU-Parallelism K-Means                 # CPU Parallel
│   ├── GPU-Parallelism K-Means                 # GPU Parallel
│   └── K-Means Benchmark                       # BenchmarkDotNet performance evaluation
│
├── Mean_Shift/                                 # Mean Shift segmentation (Algorithm)
│   ├── CPU-Parallelism Mean Shift              # CPU Parallel
│   └── GPU-Parallelism Mean Shift              # GPU Parallel
│
└── Prototype/
    └── K-Means-GPU                             # Prototype (K-Means GPU Parallel)

---

👥 Contributors

Jacob7179

jeremypangdv

t1an-wei