• Facebook
  • Yahoo
  • Twitter
  • Google
  • Live
  • Facebook
  • Yahoo
  • Twitter
  • Google
  • Live

Cytopathology using bright-field and high-resolution quantitative phase imaging of cell samples

4 Nov | By Biophotonics.World
Cytopathology using bright-field and high-resolution quantitative phase imaging of cell samples
Illustration of high resolution 3D microscope technology; (a) Microscope unit, (b) Sample cervical nuclei at different stages shown in bright-field, hologram and phase mode, (c) bright-field field of view as seen with the microscope, (d) zoomed image of a cell nucleus in bright-field mode, (e) phase map of the nucleus rendered as surface height profile. The 3D structural information available from phase image is not available from the bright-field image as used by clinicians.
By: Kedar Khare

Visual examination of cell samples using a bright-field microscope is a common methodology used world over by pathologists. A whole scan of a cell smear slide using a 40x or 60x objective amounts to tens of gigabytes of data and examining this image data in a short time is a daunting task. A number of recent studies have suggested the use of machine learning approaches to improve the cell classification tasks and at least semi-automate them. The machine learning approaches require prior labelling of cells for a training dataset. This is however not straightforward task for the early or pre-cancerous cells as there is an inter-observer variability in such cases. Our aim in this work is to consider additional physical information in the form of quantitative phase images of the cells for effective cell classification in an unsupervised manner.

Quantitative phase imaging methodology using digital holographic microscopy (DHM) is known for last couple of decades. In our opinion this methodology has not penetrated enough into the clinical usage due to the unfavourable cost to resolution trade-off with the traditional DHM systems. The single-shot off-axis DHM systems can be built at a reasonable cost but the standard Fourier filtering methodology used for phase reconstruction in such a configuration suffers from poor resolution (compared to the corresponding bright-field images) due to the inherent low-pass filtering nature of this reconstruction approach. The lost resolution may be regained in a multi-shot phase shifting approach, however, the multi-shot DHM systems require stringent vibration isolation and expensive phase-shifting hardware making the overall cost unaffordable for a pathology clinic. The team of researchers at Indian Institute of Technology Delhi (IITD) has over the past few years developed a single-shot full resolution phase reconstruction algorithm that models the phase reconstruction as a sparse optimization problem, thereby enabling building of a low-cost single-shot DHM system whose phase images have a resolution at par with the corresponding bright-field images. Since the clinicians are not used to interpret the phase images, this DHM system is configured to record a focused bright-field image of the cell sample (which the clinicians can correlate to) followed by a single-shot holographic recording with a low power laser that enables accurate phase reconstruction.

In a first study using this system in collaboration with All India Institute of Medical Sciences Delhi and a start-up Phase Laboratories incubated at IITD, our team recorded images of cervical cells from Pap-smear slides in both bright-field and high resolution phase modes. Twenty quantitative parameters were designed in consultation with pathologists and their numerical values were extracted from bright-field (RGB) and quantitative phase images of cervical cells. The parameters obtained using bright-field images (e.g. nucleus area) are already used by clinicians but the phase based parameters (e.g. optical volume) are not currently accessible to them.  A Principal Component Analysis (PCA) of this data showed that when bright-field only information was used, the different cell types were mixed with each other on the PCA plot. The cells were however organized systematically as per their cell-types when PCA analysis was performed using both bright-field as well as quantitative phase information. The result is significant in that it clearly shows the valuable non-redundant information offered by quantitative phase images. The single-shot full resolution phase imaging capability makes our DHM instrument comparable in price to a typical bright-field microscope on the one hand and also allows us to utilize fine scale textural information from phase images which appears to be important for cell classification tasks. We believe that our methodology has potential to be widely acceptable due to these features. Further since phase brings new non-redundant information about cell morphology, any machine learning approach for automated cell classification is likely to work better when information from bright-field as well as phase images is used for this purpose. 

Reference:

J. Mangal, R. Monga, S. R. Mathur, A. K. Dinda, J. Joseph, S. Ahlawat and K. Khare, “Unsupervised organization of cervical cells using bright-field and single-shot digital holographic microscopy”, Journal of Biophotonics 12, e201800409 (2019).


Author: Kedar Khare, Department of Physics, Indian Institute of Technology Delhi, New Delhi 110016 India


Share on Biophotonics.World:  
0
Share on Social Media:

Comments:

No comments
You need to sign in to comment

Related Articles

Categories

Most Viewed