LiDRoSIS: An Automated MATLAB-Python Platform for Image Processing and Quantitative Analysis of Lipid Droplets and ROS in Irradiated Cells

LiDRoSIS is an automated MATLAB-Python platform that integrates image processing and statistical analysis to enable reproducible, high-throughput quantification of lipid droplets and reactive oxygen species in irradiated cells treated with gold-based nanoparticles.

The Gist

Imagine a busy city (a cell) under stress from a storm (radiation) and some strange, shiny construction materials (gold nanoparticles) floating around. Inside this city, there are two important things scientists want to watch: fat storage bubbles (Lipid Droplets) and smoke signals (Reactive Oxygen Species, or ROS).

The problem is that looking at these tiny bubbles and smoke signals under a microscope is like trying to count raindrops in a hurricane while the camera is shaking. The images are messy, the lighting is uneven, and the bubbles often overlap or look like fuzzy clouds. Existing tools are like a manual camera that requires a human to squint, guess, and manually click on every single drop—a slow and error-prone process.

Enter LiDRoSIS.

Think of LiDRoSIS as a super-smart, two-part robot assistant that does the heavy lifting for scientists. It's built to automatically find, count, and measure these fat bubbles and smoke signals in cells that have been zapped with radiation.

Here is how it works, broken down into simple steps:

1. The "Eagle Eye" (The MATLAB Part)

The first part of the robot is like a highly trained detective with a magnifying glass. It looks at the microscope photos and does three main things:

• Finds the "City Hall": First, it locates the nucleus (the cell's control center) to know where one cell ends and another begins.
• Sorts the Fat Bubbles: It uses special filters to spot the fat bubbles. It can tell the difference between a bright, distinct bubble and a fuzzy, blurry one. It even checks if the bubbles are glowing red, green, or both (which tells scientists about the chemical state of the fat).
• Tracks the Smoke: It does the same for the "smoke signals" (ROS), distinguishing between sharp, pinpoint sparks of smoke and a general, hazy cloud of smoke.

Instead of a human guessing, this part of the software uses math to decide exactly what is a bubble and what is just background noise. It then creates a neat list of measurements for every single bubble it finds.

2. The "Data Analyst" (The Python Part)

Once the "Eagle Eye" has counted everything, it passes the list to the second part of the robot: the Data Analyst.

• Imagine the first part wrote down numbers on a spreadsheet. The Data Analyst takes that spreadsheet and instantly turns it into charts, graphs, and statistical tests.
• It answers questions like: "Did the fat bubbles get bigger when the radiation dose went up?" or "Is the smoke signal significantly stronger in the cells with gold nanoparticles?"
• It does this automatically, so the scientist doesn't have to crunch the numbers by hand.

Why is this a big deal?

The paper explains that before this tool, scientists had to do this work manually or use tools that weren't quite right for these specific, messy images.

• It's Consistent: If you run the same image through the tool ten times, you get the same answer every time. No more "human error" or tired eyes.
• It's Fast: It can process a whole folder of images in the time it takes a human to look at just one.
• It's Open: The code is free for anyone to use, look at, and tweak, much like an open-source recipe book.

The Results

The authors tested this robot on lung and breast cancer cells that were treated with gold nanoparticles and then exposed to radiation.

• The tool successfully counted the fat bubbles and measured the smoke signals.
• It proved that as the radiation dose increased, the cells showed more "smoke" (oxidative stress) and changes in their fat bubbles.
• It confirmed that the tool is sensitive enough to detect these subtle changes, which helps scientists understand how radiation and nanoparticles affect cells.

In short: LiDRoSIS is a free, automated software suite that acts like a tireless, super-accurate assistant. It takes messy microscope photos of stressed cells, automatically finds the fat bubbles and smoke signals, and turns them into clear, reliable data charts, helping scientists understand how radiation and new medical materials interact with our cells.

Technical Summary

Technical Summary: LiDRoSIS

Problem Statement
The quantification of subcellular features, specifically lipid droplets (LDs) and reactive oxygen species (ROS), in fluorescence microscopy images of irradiated cells remains a significant challenge. Existing general-purpose platforms like ImageJ, CellProfiler, and 3D Slicer often require manual intervention and lack integrated statistical post-processing. Furthermore, these tools struggle with the specific complexities of radiobiological imaging, including illumination heterogeneity, morphological variability, overlapping vesicular structures, and noise introduced by nanoparticle-induced scattering and diffuse cytoplasmic signals. Classical thresholding methods frequently fail to capture subtle morphological variations and diffuse fluorescence patterns induced by radiation and nanoparticle exposure.

Methodology
LiDRoSIS (Lipid Droplet and Reactive Oxygen Species Inference System) is an automated, hybrid software suite designed to bridge MATLAB for image processing and Python for statistical analysis. The system operates through a modular pipeline:

1. MATLAB Core (Segmentation & Quantification):

   • Preprocessing: Utilizes CLAHE (Contrast Limited Adaptive Histogram Equalization) for contrast enhancement and morphological filtering for noise removal.
   • Nuclei Segmentation: Extracts nuclear masks from the DAPI (blue) channel using adaptive Otsu thresholding to ensure biologically consistent per-cell assignment.
   • LD Detection: Employs Difference of Gaussians (DoG) and Steerable DoG (SDOG) filters to identify lipid droplets in red and green channels (Nile Red staining). It classifies droplets by polarity and computes colocalization indices. A K-means intensity clustering approach is used to detect low-contrast, diffuse lipid regions.
   • ROS Detection: Identifies both punctate and diffuse oxidative structures (DCFH-DA staining) via channel-specific intensity normalization and adaptive thresholding.
   • Feature Extraction: Calculates morphometric parameters including area, mean fluorescence intensity, circularity, eccentricity, solidity, and perimeter for each object, nucleus, and global image.
   • Output: Generates structured Excel reports containing three sheets: Global Metrics, Nucleus Metrics, and Object Metrics.

2. Python Companion (StatLysis™):

   • Imports the Excel outputs and aggregates metrics based on experimental conditions (cell line, nanoparticle type, radiation dose).
   • Performs statistical analyses including ANOVA, post-hoc tests, and linear regression.
   • Generates visualizations (boxplots, violin plots, histograms) and model-fit summaries for reproducible reporting.

The system is designed for users without programming expertise, offering a Graphical User Interface (GUI) for parameter tuning and batch processing of image folders.

Key Contributions

• Specialized Pipeline: Unlike general tools, LiDRoSIS is explicitly optimized for irradiated cellular models and fluorescence microscopy with overlapping vesicular structures, addressing the specific noise profiles of nanoparticle-enhanced radiobiology.
• Hybrid Architecture: It combines the precision of MATLAB's image processing toolbox with the statistical flexibility of Python libraries (pandas, scipy, statsmodels), ensuring a seamless workflow from raw image to statistical conclusion.
• Reproducibility and Standardization: The platform enforces standardized preprocessing and segmentation pipelines, exporting parameter-documented, structured data to facilitate cross-experiment comparisons.
• Accessibility: It operates on standard MATLAB/Python environments without requiring deep learning expertise or complex training datasets.
• Open-Source Framework: Released under the MIT License with code and documentation available on GitHub, it supports open science and educational applications in radiobiology and nanomedicine.

Results
The platform was validated using fluorescence microscopy images of A549 (human lung carcinoma) and MCF7 cells treated with gold-coated nanodiamonds (NDAuNPs) and exposed to radiation doses of 0 Gy, 2 Gy, and 10 Gy.

• Segmentation Performance: The algorithm successfully isolated vesicular LD structures while excluding diffuse cytoplasmic background and distinguished between localized ROS signals and broader diffusion patterns, outperforming conventional global thresholding in these specific contexts.
• Quantitative Findings: Analysis revealed a dose-dependent increase in LD count and fluorescence intensity in irradiated, AuNP-treated cells. The GlobalMeanRatio_Red_Green showed a significant increase with radiation dose (ANOVA, p = 0.0011), indicating LD polarity changes.
• Statistical Validation: Aggregated data confirmed significant variations in fluorescence intensity (p < 0.05) between control and irradiated groups, validating the system's sensitivity to radiation-induced oxidative stress.

Significance
The paper positions LiDRoSIS as a robust framework for the reproducible quantification of subcellular fluorescence dynamics, a critical component in radiation biology, toxicology, and nanomedicine. Its primary significance lies in filling the niche for specialized tools that handle the complex interplay of nanoparticles and ionizing radiation without relying on "black box" deep learning models. By providing an open, transparent, and parameter-controlled environment, LiDRoSIS aims to enhance scientific reproducibility in nanomedicine research and serves as an educational tool for teaching image-based quantification. The authors note that while the current version focuses on 2D fluorescence, the modular design allows for future extensions into deep learning, 3D segmentation, and other imaging modalities.