Multi-Target-Detection-in-EIT---KISMED

Project summary: Electrical impedance tomograph (EIT) utilizes non-invasive techniques for monitoring medial conditions. Its a low cost and non-invasive technology that allows for persistant imaging. However, compared to other imaging techniques, EIT has very low spatial resolution and low contrast. This results in difficulty when differentiating multiple targets within a small space. The GREIT algorithm when implemented resulted in clear images for a single object, and relatively clear images for two objects of different sizes.

Goal: What spatial and algorithmic configurations optimize the resolution of objects in a controlled tank environment?

Installation

bash git clone https://github.com/yourusername/Multi-Target-Detection-in-EIT---KISMED.git cd Multi-Target-Detection-in-EIT---KISMED pip install -r requirements.txt

Requirements

• Python 3.8+
• numpy
• matplotlib
• tqdm
• scipy
• pandas

Usage

This project extends upon a previous project linked here [https://github.com/EITatKISMED/GREIT_reconstruction]. The previous project worked idealy for single object reconstruction. However this project extends this to multi-object reconstruction. It will maintain, backwards compatability for the original project. Key note: the multi-object detection relizes on coordinate to be included as a variable within the METADATA.txt file. If this is not completed, the original sequence will run instead.

Notes

Reconstruction Techniques

• TSVD (reconstruct_images_tsvd)
• ART (reconstruct_images_art)
• SIRT (reconstruct_images_sirt)

Evaluation & Metrics Figures of Merit (FoMs):

• Amplitude Response (AR)
• Position Error (PE)
• Shape Deformation (SD)
• Ringing (RING)
• Resolution (RES)

Visualization

• plot_compare(): compare multiple reconstruction techniques
• plot_masks(): show mask overlays for detected objects
• plot_all_coordinates_mask(): visualize all ground-truth target areas

Directory

Multi-Target-Detection-in-EIT---KISMED/
├── src/
│   ├── reconstructions.py
│   ├── reconstruction_matrix.py
│   ├── MultiGround.py
│   ├── fom_multi.py
│   ├── fom.py
│   ├── utils.py
├── data/
│   ├── combined.pkl
│   ├── metadata.txt
├── results/
├── requirements.txt
└── README.md

Configuration Parameters

analyze_images()

• Creates masking behavior based on the included variables. Computes both an approximatation of number of objects and exact number of objects

      save_path,                  #Notes the location to save image results
      image_set,                 #Loops through target images from the separation function
      coordinates_t,             #Determines the error between the original coordinates and the assumed location of the objects
      label="",                  #Used to label the output figures
      image_index=1,             #Additional iteration counter
      top_n=None,                #Variable to determine if the number of objects is computed or provided
      use_filter=False,          #Determines if image should be filtered (used for TSVD images)
      use_size_based=False,      #Determines if size of the object is provided or it is assumed
      min_ratio=None,            #Determines the minimum size of objects should be detected
      reshape_shape=(32, 32),    #Shape of image
      peak_threshold_frac=0.3,   # 30% of the maximum amplitude can be included as seperate peaks
      validation_ratio=0.8,      # Determines the percentage of the object should be within 30% of the specific object peak.
  

reconstruct_images_tsvd (y_t, y_n, R, k=50, target_min=None, target_max=None)

      y_t,              #The voltage differences within the target measurement
      y_n,              #The voltage differences within the noise measurements
      R                 #The reconstruction matrix
      k=50              #Set number for determining quality of image
      target_min=None
      target_max=None