Purchase this article with an account.
Nathan Hotaling, Nicholas J Schaub, Qin Wan, Ruchi Sharma, Sarala Padi, Petre Manescu, Joe Chalfoun, Mylene Simon, Mohamed Ouladi, Carl G. Simon, Peter Bajcsy, Kapil Bharti; AMD Cell Therapy Efficacy Assessment Using Artificial Intelligence-Based Multi-Spectral Imaging. Invest. Ophthalmol. Vis. Sci. 2018;59(9):555.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop a commercially viable cell therapy product, release criteria that assess product potency, identity, and batch-to-batch variability are necessary. Here we present work showing that the function and variability of induced pluripotent stem cell derived retinal pigment epithelial cells (iPSC-RPE) from patients with age-related macular degeneration (AMD) can be assessed using novel multi-spectral imaging and convolutional neural network (CNN) deep learning algorithms.
Human iPSC-RPE were reprogrammed and differentiated using a good laboratory practices (GLP) protocol. Cells were cultured on a biodegradable poly(lactic-co-glycolic acid) nanofiber scaffold. Mature iPSC-RPE phenotype was determined using gene expression, phagocytosis of photoreceptor outer segments (POS), cytokine secretion, and transepithelial resistance (TER). In parallel to the phenotype assessment, iPSC-RPE were imaged using novel live multi-spectral acquisition and fluorescent imaging on fixed cells. Imaging of over 100,000 cells per clone at five different time points was performed. CNNs as well as traditional machine learning algorithms were then used to analyze the images and predict cell phenotype and to cluster/classify cell donors/clones.
The iPSC-RPE derived from eight clones of three donors were analyzed and found to have phagocytosed 5-15x more POS than controls, TER values >500 ohms/cm2, gene expression profiles similar to mature RPE, and demonstrated VEGF secretion that was polarized. When combined, multi-spectral imaging and CNNs accurately predicted cell function with high sensitivity and specificity as well as classified clones and donors to an equal degree as that of traditional physiological and molecular assays. The CNN networks and machine learning algorithms were able to determine cell function using multiple approaches and visual parameters that are important for predicting function were identified.
Using only non-invasive multi-spectral visual data, CNNs, and machine learning we were able to assess cell function and clustering to an equal degree as that of traditional molecular and physiological assays. This process is fully automated, relatively inexpensive, and needs no human intervention to perform. This approach allows for the creation of a release criterion for an iPSC-RPE cell based therapy that may be useful in manufacturing or in a clinical setting.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.
This PDF is available to Subscribers Only