Hyperspectral Imaging – Optical Modelling for Histological Image Segmentation

Project Overview

Hyperspectral imaging (HSI) captures optical infromation - rich spectral information across a wide range of wavelengths, enabling the analysis of biological tissue without the need for invasive sampling.

This project implements a physics-based simulation of hyperspectral imaging for tissue segmentation in histological slides. By modelling the optical absorption characteristics of haemoglobin, the method enables segmentation driven by spectral (optical) information rather than colour or texture. The pipeline was developed in MATLAB and applies principles from biomedical optics and computational imaging.

🛠 Methods & Techniques

This project involved a complete simulation and segmentation pipeline, structured around the following stages:

🔬 Optical Simulation

• Modelled the light–tissue interaction using oxyhaemoglobin and deoxyhaemoglobin absorption spectra.
• Applied the Modified Beer–Lambert Law to simulate pixel-level absorbance across wavelengths: $$ A(\lambda) = \varepsilon_{\mathrm{HbO_2}}(\lambda) \cdot c_{\mathrm{HbO_2}} \cdot L + \varepsilon_{\mathrm{Hb}}(\lambda) \cdot c_{\mathrm{Hb}} \cdot L $$
• Simulated reflectance under standard daylight illumination (D65) to mimic real imaging conditions.

🎨 RGB & Spectral Reconstruction

• Converted the spectral absorbance data into reflectance using physics-based modelling.
• Interpolated narrowband data to generate RGB representations of the tissue.
• Created a simulated histology image showing tissue contrast based on absorption differences rather than just stain colour.

🧩 Pixel-wise Spectral Modelling

• Analysed reflectance per pixel to construct a spectral signature for each region.
• Used these signatures to detect subtle biochemical variations within the tissue.

🧠 Segmentation

• Identified regions with varying oxygenation levels through contrast in spectral absorption.
• Segmented tissue structures using thresholds or clustering over the simulated spectral domain.

📸 Result Preview

The resulting segmented image highlights tissue structures by exploiting variations in spectral absorption, showing how hyperspectral modelling can reveal underlying biological properties invisible in standard RGB images.

🧰 Technical Highlights

• MATLAB programming for image processing and simulation
• Application of optical-based modelling in computer vision
• Image reconstruction using Spectral data interpolation and RGB reconstruction
• Tissue segmentation using wavelength-dependent absorption modelling
• Data visualisation and interpretation of simulated histological contrast

Practical Relevance

This project demonstrates the application of physics-based modelling to enhance tissue segmentation in medical imaging. By simulating light–tissue interactions and extracting biologically meaningful features from spectral absorption profiles, it showcases the integration of:

• Domain-specific biomedical knowledge with computational imaging
• Optical modelling and spectral analysis for vision-based tasks
• Simulation-driven approaches to segmentation and tissue characterisation
• Numerical data processing for image reconstruction and analysis

The techniques used are directly applicable to problems in biomedical computer vision, spectral image analysis, and interpretable modelling for healthcare imaging.