Research Methodology: Bacterial Growth Optical Flow Analysis

1. Project Overview

This repository hosts a high-throughput computer vision pipeline designed to quantify bacterial phenotypic responses to antibiotic stress. By replacing computationally expensive deep learning segmentation (O(N)) with dense optical flow estimation (O(1)), this system provides a lightweight, real-time kinetic proxy for bacterial mortality.

The pipeline utilizes Farneback’s algorithm to compute dense vector fields, applying statistical outlier rejection to isolate biological motility from stochastic sensor noise. It successfully identifies the "Point of Divergence" between Rifampicin-treated (RIF) and Control (REF), offering a computationally efficient, zero-latency alternative to traditional biomass area integration methods.

Key Engineering Features

• Computer Vision: Dense optical flow estimation (Farneback) with spatial smoothing kernels to track coherent colony expansion.
• Signal Processing: Double-threshold filtering (Bottom 50% / Top 5%) to eliminate Brownian motion noise and camera artifacts.
• Event Detection: Automated temporal gating to identify pharmacological injection events ("fluid dynamic perturbations") and exclude them from growth metrics.

2. Directory Structure

The codebase is organized into a modular package structure to separate business logic from execution scripts.

.
├── data/                          # Raw experimental image stacks
│   ├── REF_raw_data101_110/       # Control Group (Untreated)
│   └── RIF10_raw_data201_210/     # Experimental Group (Treated)
├── results/                       # Generated plots and time-series data
├── src/                           # Source Code
│   ├── optical_flow_pipeline.py   # Main differential analysis engine
│   ├── injection_event_detector.py# Temporal perturbation monitor
│   └── vector_field_visualizer.py # Qualitative heatmap generator
├── requirements.txt               # Dependencies
└── README.md

3. Setup & Installation

Dependencies

Install the required Python libraries:

pip install -r requirements.txt

Output: Generates results_comparison.png, showing the net flow difference between REF and RIF and marking the exact frame where behavior diverges.

Data Preparation

Ensure raw data is placed in the data/ directory as shown in the structure above.

4. Usage Instructions

Method 1: Quantitative Differential Analysis

Executes the core pipeline to generate kinetic time-series comparisons and statistical divergence plots.

python src/optical_flow_pipeline.py

Output: Generates results/results_comparison.png, visualizing the net flow difference and statistically identifying the onset of necrotic cessation.

Method 2: Injection Event Detection

Analyzes temporal vector magnitude spikes to pinpoint the exact frame of compound administration.

python python src/injection_event_detector.py

Output: Generates a temporal profile identifying the "Injection Window" (fluid perturbation artifact) to ensure downstream data integrity.

Method 3: Vector Field Visualization

Generates optical vector filed to visually verify flow dynamics and directional bias.

python python src/vector_field_visualizer.py

Output: Generates results/results_comparison.png, visualizing the net flow difference and statistically identifying the onset of necrotic cessation.

5. Interpretation of Results

The results from this method are subtle. The antibiotic effect does not cause the bacteria to immediately disappear; rather, it biases their movement direction.

A. Quantitative Analysis (results/results_comparison.png)

This image contains 5 panels. Here is how to read them:

1. Horizontal Flow (Top Panel):

   • What it shows: The raw movement trajectories of Untreated (REF, Blue) vs. Treated (RIF, Red).
   • Interpretation: You may see the lines overlapping significantly. The signals of separation here are not always clear to the naked eye because both colonies are growing.

2. Vertical Flow (Second Panel):

   • What it shows: Up/Down movement.
   • Interpretation: Generally serves as a control. Both conditions typically oscillate around 0 (no significant vertical bias), confirming that the chamber setup is level.

3. Flow Magnitude (Third Panel):

   • What it shows: The overall "intensity" of movement, regardless of direction.
   • Interpretation: Treated bacteria often show higher magnitude spikes due to stress responses, but this signal can be noisy.

4. Difference Plot (Fourth Panel - Most Important):

   • What it shows: The mathematical difference: RIF (Treated) - REF (Untreated).
   • Interpretation: This is where the signal becomes visible. You will likely see positive values, indicating that the RIF-treated bacteria have a persistent rightward inclination compared to the control. Even if the signal is weak, a consistent positive trend indicates the "declining" or dying phase where biomass drifts or expands asymmetrically.

5. Statistical Summary (Bottom Right):

   • Provides the global averages. If RIF10 Avg Horizontal is higher than REF, it confirms the rightward directional bias quantitatively.

B. Qualitative Visualization (results/)

• Files: Images like viz_RIF_frame030.png.
• Interpretation: Do not look for a difference in the "size" or "area" of the colony, as both grow initially.
• What to look for: Look at the arrows (Vectors).
  • REF (Untreated): Arrows point outward in all directions (healthy symmetric expansion).
  • RIF (Treated): You may see a tendency for arrows to align rightward (yellow/red arrows). This directional bias is the visual signature of the antibiotic treatment taking effect.