Automated pancreatic islet viability assessment for transplantation using bright-field deep morphological signature

Abbas Habibalahi; Jared M Campbell; Stacey N Walters; Saabah B Mahbub; Ayad G Anwer; Shane T Grey; Ewa M Goldys

doi:10.1016/j.csbj.2023.02.039

Automated pancreatic islet viability assessment for transplantation using bright-field deep morphological signature

Comput Struct Biotechnol J. 2023 Feb 24:21:1851-1859. doi: 10.1016/j.csbj.2023.02.039. eCollection 2023.

Authors

Abbas Habibalahi^{1

2}, Jared M Campbell^{1

2}, Stacey N Walters^{3

4

5}, Saabah B Mahbub^{1

2}, Ayad G Anwer^{1

2}, Shane T Grey^{3

4

5}, Ewa M Goldys¹

Affiliations

¹ Graduate School of Biomedical Engineering, University of New South Wales, Sydney, NSW 2052, Australia.
² Australian Research Council Centre of Excellence for Nanoscale BioPhotonics, Australia.
³ Garvan Institute of Medical Research, Darlinghurst, New South Wales, Australia.
⁴ Faculty of Medicine, University of New South Wales, Sydney, NSW, Australia.
⁵ St Vincent's Clinical School, The University of New South Wales, Sydney, NSW, 2010 Australia.

Abstract

Islets transplanted for type-1 diabetes have their viability reduced by warm ischemia, dimethyloxalylglycine (DMOG; hypoxia model), oxidative stress and cytokine injury. This results in frequent transplant failures and the major burden of patients having to undergo multiple rounds of treatment for insulin independence. Presently there is no reliable measure to assess islet preparation viability prior to clinical transplantation. We investigated deep morphological signatures (DMS) for detecting the exposure of islets to viability compromising insults from brightfield images. Accuracies ranged from 98 % to 68 % for; ROS damage, pro-inflammatory cytokines, warm ischemia and DMOG. When islets were disaggregated to single cells to enable higher throughput data collection, good accuracy was still obtained (83-71 %). Encapsulation of islets reduced accuracy for cytokine exposure, but it was still high (78 %). Unsupervised modelling of the DMS for islet preparations transplanted into a syngeneic mouse model was able to predict whether or not they would restore glucose control with 100 % accuracy. Our strategy for constructing DMS' is effective for the assessment of islet pre-transplant viability. If translated into the clinic, standard equipment could be used to prospectively identify non-functional islet preparations unable to contribute to the restoration of glucose control and reduce the burden of unsuccessful treatments.

Keywords: AI, artificial intelligence; DMOG, dimethyloxalylglycine; DMS, deep morphological signatures; Deep morphological signature; ECG, electrocardiogram; EEG, electroencephalogram; EMCCD, electron multiplying charge coupling device; FD, Fisher Distance; GSIS, glucose stimulated insulin secretion; IoU, intersection over union; MEG, magnetoencephalography; MRI, magnetic resonance imaging; PCA, principal component analysis; Pancreatic islet; ROS, reactive oxygen species; SI, swarm intelligence; SVM, support vector machine; Transplantation.