Intelligent Digital Inline Holographic Micrograph (DIHM) Cell-Enhancement and Characterization

Rajkumar Vaghashiya, Dr. Mohendra Roy [(Assistant Professor, PDPU)](https://sites.google.com/view/mohendraroylab/home)
Aug 1, 2019 2 min read Demo

Intelligent Digital Inline Holographic Micrograph (DIHM) Cell-Enhancement and Characterization

Project Intro

Developing a novel and efficient neural model for detecting and classifying cells such as RBC, WBC, and cancer cells, etc.(cellular pathology) from DIH(Digital Inline Holographic) microscopy images

Implementation on a resource-constrained device to develop a cheap, reliable and portable point-of-care testing facility for diagnosis of pathological diseases, especially for usage in rural areas

Highlights

  • Segment cell-lines in DIH micrograph; performs signal enhancement using CNN-based autoencoder, followed by the cell-line characterization
  • ROC-AUC: >0.98 for RBC, WBC, and microbeads; >0.88 for cancer cells HepG2 and MCF7
  • Easy accommodation of newer cell-lines. Python, TensorFlow, OpenCV
  • Preliminary work published at the 8th IEEE ICHI
Breast Cancer Disease Statistics
Breast Cancer Disease Statistics
Limitations of Traditional Optic Microscopy
Limitations of Traditional Optic Microscopy

Aim

The aim of the research is to create a

  • Cheap,
  • Reliable,
  • Adaptable,
  • Portable, and
  • Intelligent real-time point-of-care testing facility that could be used in resource-constrained environments, especially such as those in a rural setting.

Dataset

Cell-lines used in the project
Cell-lines used in the project

Current Progress

Cell-line EDA
Cell-line EDA

The colored cell-lines is the result of EDA on the statistical properties and pixel intensity in the images

Segmentation Results

The segmentation involves bit-plane splicing, adaptive thresholding, and contour approximation to crop the cell-lines
The segmentation involves bit-plane splicing, adaptive thresholding, and contour approximation to crop the cell-lines

CNN Model Performance

Confusion Matrix
Confusion Matrix

The recognition performance of the CNN model on the augmented dataset

Final Cell Counts for the Input Micrograph

The final result in form of cell-counts for the ROI (region of interest) window selected in the DIH micrograph
The final result in form of cell-counts for the ROI (region of interest) window selected in the DIH micrograph
Academic Research Deep Learning Pathology
Rajkumar Vaghashiya
Rajkumar Vaghashiya
MS in Computer Science

My research interests include applied machine learning and computer vision.

Related