Automatic detection and characterization of quantitative phase images of thalassemic red blood cells using a mask region-based convolutional neural network.

Yang-Hsien Lin,Ken Y-K Liao,Kung-Bin Sung

doi:10.1117/1.jbo.25.11.116502

Yang-Hsien Lin, Ken Y-K Liao + Show 1 more

Open Access

https://doi.org/10.1117/1.jbo.25.11.116502

Copy DOI

Abstract

.Significance: Label-free quantitative phase imaging is a promising technique for the automatic detection of abnormal red blood cells (RBCs) in real time. Although deep-learning techniques can accurately detect abnormal RBCs from quantitative phase images efficiently, their applications in diagnostic testing are limited by the lack of transparency. More interpretable results such as morphological and biochemical characteristics of individual RBCs are highly desirable.Aim: An end-to-end deep-learning model was developed to efficiently discriminate thalassemic RBCs (tRBCs) from healthy RBCs (hRBCs) in quantitative phase images and segment RBCs for single-cell characterization.Approach: Two-dimensional quantitative phase images of hRBCs and tRBCs were acquired using digital holographic microscopy. A mask region-based convolutional neural network (Mask R-CNN) model was trained to discriminate tRBCs and segment individual RBCs. Characterization of tRBCs was achieved utilizing SHapley Additive exPlanation analysis and canonical correlation analysis on automatically segmented RBC phase images.Results: The implemented model achieved 97.8% accuracy in detecting tRBCs. Phase-shift statistics showed the highest influence on the correct classification of tRBCs. Associations between the phase-shift features and three-dimensional morphological features were revealed.Conclusions: The implemented Mask R-CNN model accurately identified tRBCs and segmented RBCs to provide single-RBC characterization, which has the potential to aid clinical decision-making.

Highlights

Information about red blood cell (RBC) morphology is crucial to reach a diagnosis of blood disorders because red blood cells (RBCs) morphologies often change due to altered membrane lipid composition, iron deficiency, or metabolic abnormalities.[1,2] The current diagnostic procedure for many blood-related diseases is the blood smear test in which the morphology of stained RBCs is examined under a light microscope
We demonstrate the implementation of Mask R-convolutional neural networks (CNNs) for the automatic discrimination between quantitative phase images of healthy RBCs (hRBCs) and those of thalassemic RBCs (tRBCs)
Our findings suggest the use of quantitative phase imaging (QPI) as a useful tool to capture 3-D morphological characteristics of abnormal RBCs, such as those seen in patients with thalassemia

Summary

Introduction

Information about red blood cell (RBC) morphology is crucial to reach a diagnosis of blood disorders because RBC morphologies often change due to altered membrane lipid composition, iron deficiency, or metabolic abnormalities.[1,2] The current diagnostic procedure for many blood-related diseases is the blood smear test in which the morphology of stained RBCs is examined under a light microscope. The optical volume has been shown to be a major feature in identifying hypochromic RBCs, as seen in iron deficiency anemia[9,10] and thalassemia.[10,11] In addition, phase-shift statistics and two-dimensional (2-D) morphological features, such as the projected area, perimeter, and lengths of major and minor axes, have been extracted from quantitative phase images of RBCs and have been used to classify blood disorders, including hereditary spherocytosis,[4,9] malaria,[12,13,14] and sickle cell disease,[15,16,17] where RBCs possess abnormal hemoglobin content and show abnormal shape These advances highlight a major advantage of using QPI for the diagnosis of RBC-related diseases: simultaneously quantifying diagnostically relevant hemoglobin content and morphological information without exogenous labels

Objectives

Methods

Results

Discussion

Conclusion