ClearAIM

⚠️ This repository has been moved!

📦 The project is now maintained and developed under the QCI-LAB organization.

👉 New repository URL: https://github.com/QCI-LAB/ClearAIM

Overview

This repository contains scripts for image processing and analysis, specifically for detecting and analyzing brain images. The main functionalities include mask prediction using the Segment Anything Model (SAM) and subsequent analysis of the predicted masks.

ClearAIM stands for Clear Artificial Intelligence Monitoring, emphasizing the focus on using AI techniques for precise and transparent analysis in biomedical imaging.

Requirements

Python Script (main_mask_detection.py)

• Python 3.6+
• PyTorch with CUDA support
• OpenCV
• NumPy
• Matplotlib
• tqdm

MATLAB Script (main_analize.m)

• MATLAB R2018b or later
• Image Processing Toolbox

Installation

Python Environment

1. Install PyTorch with CUDA:

   pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117

2. Install other dependencies:

   pip install opencv-python numpy matplotlib tqdm

MATLAB Environment

Ensure you have MATLAB installed with the Image Processing Toolbox.

Usage

Python Script (main_mask_detection.py)

This script uses the Segment Anything Model (SAM) to predict masks for images in the Materials directory and saves the masks in the Result directory.

1. Prepare the environment:

   • Ensure you have the SAM model checkpoint file (sam_vit_h.pth) in the models directory.

2. Run the script:

   python main_mask_detection.py

3. Script Workflow:

   • Loads the SAM model.
   • Processes each image in the Materials directory.
   • Predicts the mask for each image.
   • Saves the mask in the Result directory.

MATLAB Script (main_analize.m)

This script analyzes images and their corresponding masks, calculates measurements such as Weber contrast and area, and generates visualizations.

1. Prepare the environment:

   • Ensure you have images in the Materials directory and corresponding masks in the Results directory.

2. Run the script:

   • Open MATLAB.
   • Navigate to the directory containing main_analize.m.
   • Run the script:

     main_analize

3. Script Workflow:

   • Finds and sorts image-mask pairs.
   • Displays images with mask contours.
   • Calculates and displays measurements such as Weber contrast and area.
   • Generates plots for the measurements.
   • Creates an animation showing the processing progress.

Functions

Python Script (main_mask_detection.py)

• SamPredictorWrapper: A wrapper class for the SAM model.
  • __init__: Initializes the SAM wrapper with model type, checkpoint path, and device.
  • _load_model: Loads the SAM model based on the specified model type and checkpoint path.
  • _prepare_points: Prepares and processes positive and negative points for further use.
  • predict_mask: Predicts a mask for the given RGB image based on positive and negative points.
• ImagePathUtility: Utility class for handling image paths.
  • get_image_paths: Retrieves image paths from the specified directory.
  • save_mask_as_image: Saves the logical mask as an image at the given path.
• ImageProcessor: Class for processing images.
  • load_image: Loads an image from the specified path.
  • rescale: Rescales the image by the given factor.
  • invert_mask: Inverts the given mask.
  • get_biggest_object_from_mask: Retrieves the biggest object from the mask.
• MaskDetector: Main class for detecting masks.
  • process_images: Main processing pipeline for images.
  • process_single_image: Processes a single image.
  • _get_save_path: Generates the save path for the mask.

MATLAB Script (main_analize.m)

• findImageMaskPairs: Finds pairs of image and mask files.
• processMask: Processes the mask by converting it to grayscale, resizing, and binarizing.
• displayImageWithMaskContour: Displays an image with overlaid mask contours.
• weberContrast: Calculates the Weber contrast.
• sort_image_mask_struct: Sorts a structure array based on numerical values in the image path.
• calcMetricsOfAllImages: Calculates metrics for all images.
• displayMetrics: Displays calculated metrics.
• createAnimationOfObjectDetection: Creates an animation showing the object detection process.

Example Usage

Python Script

from src.mask_detector import MaskDetectorBuilder

def transform_source_path_to_save_path(path_source: str) -> str:
    return path_source.replace("Materials", "Result")

if __name__ == "__main__":
    builder = MaskDetectorBuilder()
    builder.folderpath_source = r".\Materials\250um brain\skrawek 2"
    builder.folderpath_save = transform_source_path_to_save_path(builder.folderpath_source)
    builder.is_display = True
    
    detector = builder.build()
    detector.process_images()