Image-Based Cell Quality Assessment: Modeling of Cell Morphology and Quality for Clinical Cell Therapy - Computational Modeling in Tissue Engineering - page 203

measures from each morphological feature were tagged with the teacher signal
(the target prediction value determined by the experimental data collection) and
were applied as a dataset. For the modeling, we chose regression analysis models,
discriminant analysis models, clustering analysis models, or neural network-based
models, according to the complexity and quality of the teacher signal measure-
ments. Further in this chapter, we present some of our successful models.
3 Regression Analysis Model for Image-Based Cell
Quality Assessment
Regression is a modeling approach to understand the quantitative relationship
between multiple independent variables (input features) and a dependent variable
(target prediction value). During the process of constructing the regression func-
tion, users can estimate the combinational importance of various features such as
''morphological features'' in the case of image-based assessment. In other words,
users can quantitatively understand the characteristic morphological parameter
combinations that cell-culture experts unconsciously recognize and apply for their
judgements.
For the target prediction value, there are various biological parameters that require
prediction in clinical tissue engineering, such as cellular activity, cellular prolifer-
ation rate, cellular lineage, cellular differentiation rate, cellular production rate.
Among the many candidate parameters, we chose one of the essential parameters in
the cell production process, the future cell yield. The cell yield greatly affects the
scheduling of operations, because most medical facilities assure the quality of cell
therapy by defining the ''cell number for injection.'' Therefore, cell culture experts
are required to predict the operation date based on their expertise.
To replace such ambiguous cell production procedures, we sought to quantitatively
predict the future cell yield (14 days later) of clinically obtained primary human
dermal fibroblasts with multiple regression models using early cellular images (images
from 1 to 3 days culture period). Fibroblasts are the cell source for skin defect and
wrinkle medications already used in clinical cell therapy [ 2 , 25 ]. The concept of the cell
yield prediction model using cellular images is illustrated in Fig. 3 .
To obtain input features from cell culture images, we collected a total of
270 phase contrast microscopic images (20 9) of cultured primary dermal fibro-
blasts obtained from ten healthy volunteers (3 males, 7 females, 29-72 years old).
Informed consent was obtained according to a protocol approved by the Ethics
Committee of Nagoya University Hospital. Cells derived from passages 3 and 4 of
the primary expansion were maintained in modified Eagle's medium (DMEM, Life
technologies, Carlsbad, CA, USA) containing 10% fetal bovine serum (FBS, Life
technologies) at 37C in the presence of 5% CO 2 . All images (.bmp files),
manually obtained by three operators in this work, were processed using
MetaMorph software (Molecular Devices, LLC., Sunnyvale, CA, USA) with