This repository explores how MedSAM (Medical Segment Anything Model) can be adapted for underwater sonar imaging, specifically targeting fish segmentation using the Caltech Fish Counting (CFC) dataset. By leveraging the similarities between ultrasound and sonar images—particularly their noise characteristics and diffused boundaries—this project aims to demonstrate MedSAM’s potential for robust cross-domain generalization.

• project_ultrasound_to_underwater.ipynb
  An interactive Jupyter Notebook that walks through the main pipeline, including data loading, preprocessing, segmentation via MedSAM, and preliminary results analysis.

• histogram.py
  Contains functions to compute and compare grayscale histograms. Metrics such as correlation, Chi-Square distance, intersection, and Bhattacharyya distance quantify the similarity between ultrasound and sonar images.

• nps.py
  Provides utilities to compute the Noise Power Spectrum (NPS), a measure of how noise energy is distributed across spatial frequencies. Helpful for examining the parallel noise profiles in ultrasound and sonar data.

• speckle.py
  Implements speckle noise analysis, including speckle contrast (standard deviation over mean) and Shannon entropy calculations. These metrics illustrate how sonar images exhibit noise traits akin to ultrasound, guiding potential noise-reduction strategies.

• dsa.py
  Demonstrates Digital Subtraction Angiography (DSA) for motion detection via frame differencing.
  Example: python dsa.py shows baseline, contrast, and subtraction results.

• registration.py
  Implements deformable image registration using:

  • Control point detection via Canny edges
  • Local displacement estimation
  • Triangular mesh warping (Delaunay triangulation + affine transforms)
    Usage: Aligns sequential sonar frames for improved subtraction.

• non_rigid_registration.py
  Advanced registration pipeline featuring:

  • Morphological bottom-hat preprocessing
  • Hybrid edge/corner control points
  • Entropy-based displacement estimation
  • RBF-interpolated deformation fields
    Usage: Handles complex non-rigid motions in underwater scenes.

1. Clone this repository:

   git clone https://github.com/<username>/UltrasoundToUnderwater.git
   cd UltrasoundToUnderwater

2. Install dependencies (preferably in a virtual environment or conda environment):

   pip install -r requirements.txt

   • Make sure you have Python 3.7+ and a suitable version of PyTorch if you plan to integrate with MedSAM.

3. Run the notebook for a step-by-step demo:

   jupyter notebook project_ultrasound_to_underwater.ipynb

   This notebook showcases MedSAM segmentation results on example Caltech fish sonar images.

4. Scripts:

   • histogram.py:

     python histogram.py

     Analyze histograms and their similarities between images.
   • nps.py:

     python nps.py

     Visualizes and compares noise power spectra between images.
   • speckle.py:

     python speckle.py

     Prints out speckle contrast and image entropy.
   • dsa.py:

     python dsa.py

     Displays side-by-side subtraction of sonar frames to isolate fish objects.

1. Cross-Domain Segmentation: Demonstrate how MedSAM, despite being trained on medical data, can be repurposed for sonar imagery by exploiting parallels with ultrasound.
2. Noise Characterization: Show the resemblance in noise power spectrum and speckle properties between ultrasound and sonar images.
3. Enhancement Techniques: Evaluate DSA-inspired subtraction as a preprocessing step to improve fish segmentation accuracy.

• Integration with MedSAM: Further refine prompts and bounding boxes for sonar images to address over-segmentation issues.
• Extended Datasets: Explore additional sonar datasets, or incorporate more diverse ultrasound samples to bolster cross-domain evidence.
• Noise Reduction: Investigate advanced filters or deep-learning-based denoising methods to complement NPS and speckle analyses.

• MedSAM
  Ma, J., He, Y., Li, F., Han, L., You, C., & Wang, B. (2024). Segment Anything in Medical Images. Nature Communications, 15(1), 654.

• Caltech Fish Counting Dataset
  Kay, J., Kulits, P., Stathatos, S., Deng, S., Young, E., Beery, S., ... & Perona, P. (2022). The Caltech Fish Counting dataset: a benchmark for multiple-object tracking and counting. In European Conference on Computer Vision (pp. 290-311). Cham: Springer.

• Registration Methods
  Nejati, M., Sadri, S., & Amirfattahi, R. (2013). Nonrigid Image Registration in Digital Subtraction Angiography Using Multilevel B-Spline. BioMed Research International.