A weakly supervised deep learning approach for label-free imaging flow-cytometry-based blood diagnostics

Corin F Otesteanu,Martina Ugrinic,Gregor Holzner,Yun-Tsan Chang,Christina Fassnacht,Emmanuella Guenova,Stavros Stavrakis,Andrew Demello,Manfred Claassen

doi:10.1016/j.crmeth.2021.100094

Abstract

SummaryThe application of machine learning approaches to imaging flow cytometry (IFC) data has the potential to transform the diagnosis of hematological diseases. However, the need for manually labeled single-cell images for machine learning model training has severely limited its clinical application. To address this, we present iCellCnn, a weakly supervised deep learning approach for label-free IFC-based blood diagnostics. We demonstrate the capability of iCellCnn to achieve diagnosis of Sézary syndrome (SS) from patient samples on the basis of bright-field IFC images of T cells obtained after fluorescence-activated cell sorting of human peripheral blood mononuclear cell specimens. With a sample size of four healthy donors and five SS patients, iCellCnn achieved a 100% classification accuracy. As iCellCnn is not restricted to the diagnosis of SS, we expect such weakly supervised approaches to tap the diagnostic potential of IFC by providing automatic data-driven diagnosis of diseases with so-far unknown morphological manifestations.

Highlights

The accurate and sensitive diagnosis of pathologies is an essential determinant of patient treatment outcome and prognosis.Given that cell morphology, structure, and chemical composition are linked to physiological function, they can be used as essential markers for diagnosis (Alizadeh et al, 2020)
In contrast to conventional strongly supervised approaches, which are tedious and whose establishment requires individual cell images labeled as ‘‘specific’’ or ‘‘unspecific’’ to the disease, iCellCnn can be trained by using only the information on the disease state at the level of the specimen, i.e., the entire cell image collection that results from the specimen. iCellCnn circumvents the requirement for strong supervision by using a set of images rather than individual images as an input, in a similar fashion as reported for conventional flow cytometry data (Arvaniti and Claassen, 2017)
In contrast to the comparably low dimensionality of the conventional flow cytometry input (Arvaniti and Claassen, 2017), our approach employs highdimensional imaging flow cytometry (IFC) single-cell images summarized by a vector of relevant morphology features defined in a data-driven fashion (Goodfellow et al, 2020)

Summary

Introduction

Structure, and chemical composition are linked to physiological function, they can be used as essential markers for diagnosis (Alizadeh et al, 2020). Imaging flow cytometry (IFC) has emerged as a powerful tool for high-throughput single-cell morphology analysis and, in conjunction with machine learning approaches, has the potential to transform diagnosis of hematological diseases (Doan et al., 2018). Such diagnostic procedures rely on manual expert microscopical evaluation of blood cell morphology and suffer from subjectivity, limited throughput, and low sensitivity. To circumvent the problems related to the requirement for labels when identifying molecular diagnostic markers, IFC can provide high-resolution morphological information of individual cells at

Methods

Results

Discussion

Conclusion