Medical Image Analysis

Level-set segmentation and rigid registration of medical DICOM images in MATLAB.

Overview

This repository implements a medical image analysis pipeline with two modules:

• Segmentation -- Chan-Vese and Malladi-Sethian level-set methods applied to four clinical scenarios (kidney, cardiac ventricles, left atrium in 3D, breast lesion).
• Registration -- Exhaustive-search rigid registration (translation and rotation) evaluated with three similarity metrics (NCC, SSD, MI).

All analysis scripts share a common library of 20+ custom MATLAB functions covering numerical operators, similarity metrics, preprocessing, and publication-quality visualization.

Algorithms Implemented

Segmentation

Chan-Vese Active Contours -- Solves the piecewise-constant Mumford-Shah functional via level-set evolution. The contour minimizes a combined energy of curvature regularization and inside/outside intensity fitting. Used for kidney, medulla, and left atrium segmentation.

Malladi-Sethian Level Sets -- Geometric PDE-based segmentation using an edge indicator function g = 1/(1 + |nabla I / beta|^alpha). The level set evolves under curvature, advection, and edge-stopping forces. Used for ventricle and breast lesion segmentation.

Registration

Translational Registration -- 2D exhaustive grid search over a discrete shift range. Each candidate translation is scored with NCC (maximize), SSD (minimize), or MI (maximize). Applied to multimodal brain imaging (T2, DWI, PET).

Rotational Registration -- Angular sweep from 0 to 359.5 degrees at 0.5-degree resolution, using the same three metrics. Applied to rotated MRI alignment.

Case Studies

Script	Clinical Application	Algorithm	Input Modality
seg_kidney_medulla	Kidney + medulla segmentation	Chan-Vese (with pre-registration)	Contrast / non-contrast CT
seg_ventricles	Left & right ventricle endocardium	Malladi-Sethian	Cardiac MRI
seg_left_atrium_3d	3D left atrium volume + mesh	Chan-Vese (slice-by-slice) + iso2mesh	3D MRI volume
seg_breast_lesion	Breast lesion delineation	Malladi-Sethian	Breast MRI
reg_translation	Multimodal brain registration	Exhaustive translation search	T2 / DWI / PET
reg_rotation	Rotation correction	Exhaustive angle sweep	T2 MRI

Repository Structure

medical-image-analysis/
├── setup.m                        # Path initialization (run this first)
├── scripts/
│   ├── main_gui.m                 # GUI launcher
│   ├── segmentation/              # 4 segmentation case studies
│   └── registration/              # 2 registration case studies
├── lib/
│   ├── core/                      # Segmentation engines, registration framework, preprocessing
│   ├── operators/                 # Finite difference operators (Dx, Dy, Laplacian, etc.)
│   ├── metrics/                   # Similarity metrics (NCC, SSD, MI, entropy)
│   └── visualization/             # Plotting and reporting functions
├── data/                          # DICOM images and 3D MRI volume
├── results/                       # Saved figures and metrics (generated at runtime)
└── docs/                          # Documentation and portfolio assets

See CONTRIBUTIONS.md for a detailed mapping of which code is original, which is adapted from course materials, and which is third-party.

Getting Started

Requirements

• MATLAB R2020b or later
• Image Processing Toolbox
• iso2mesh v1.9.6 (required only for seg_left_atrium_3d)

Setup

% 1. Clone the repository
% 2. Open MATLAB and navigate to the project root
% 3. Initialize paths:
setup

% 4. (Optional) Install iso2mesh for 3D segmentation:
%    Download from http://iso2mesh.sf.net and add to MATLAB path

Running

GUI launcher:

main_gui

Individual scripts:

setup                    % initialize paths (once per session)
seg_kidney_medulla       % or any other script name

The registration scripts (reg_translation, reg_rotation) run fully automatically. The segmentation scripts require interactive seed-point selection via mouse clicks.

Custom Function Reference

Function	Category	Description
runChanVese	Core	Chan-Vese level-set evolution with convergence detection
computeLSF	Core	Level-set initialization with interactive seed selection
smartRegister	Core	Unified registration framework (translation / rotation)
anisodiff2D	Core	Perona-Malik anisotropic diffusion filter
NCC, SSD, MI	Metrics	Normalized cross-correlation, sum of squared differences, mutual information
plotLSF	Visualization	3D surface + 2D contour level-set display
plotEvolution	Visualization	Animated segmentation evolution with area tracking
generate_registration_report	Visualization	Multi-panel registration quality dashboard

License

This project is licensed under the MIT License. See LICENSE for details.

The iso2mesh toolbox (not bundled) is licensed under GPL v2+.