Biomedical Engineering Reference
In-Depth Information
be measured quantitatively. For example, bacterial contamination is the most well
known risk in the cell culture process. The contamination can be measured by
colony-forming assays or quantitative polymerase chain reaction (PCR) of bac-
terial markers. However, the quantity of contamination is meaningless in a model,
because any amount of contamination should be eliminated from the process. In
such cases, a decision should be made as ''yes, OK to continue'' or ''no, discard
the sample,'' and simple discriminant function models are effective in incorpo-
rating combinational features in the binary decision with high accuracy.
For discriminant function analysis, we designed a model to detect ''human
processing error'', because from the aspect of establishing more safe cell pro-
duction process, detection of ''processing error'' is essential. Practically, we
attempted to model various types of human error that could be involved in the
culture process and considered to affect the cellular damage rate. In various
studies, we succeeded in constructing a model to detect the ''error in trypsin
treatment.'' Trypsin treatment is an essential step in subculture to digest the cel-
lular adhesion molecule and collect cells from the culture plate, although known as
a physical stress that damage cells. Some delicate cells are known to be very
sensitive to trypsin concentration, and in such cases, a very low trypsin concen-
tration is recommended. Commonly, protocol of trypsin treatment is fixed in the
medical facilities' standard protocols. However, if the human error in diluting such
a damageable solution can be non-destructively monitored, the model can assist
the rigid protocol in the cell production process, and can reduce costly error-
monitoring process of additional manual checks. The concept of the human error
detection model using cellular images is illustrated in Fig. 5 .
To obtain input features from cell culture images, we collected a total of
450 phase contrast microscopic images (109) [3 wells 9 5 fields 9 15 time points
(24-136 h, every 8 h) 9 2 treatment conditions (0.25%, 0.025%)] of primary
human gingival fibroblasts obtained from healthy volunteers (22 and 24 years old,
female and male) and cultured in DMEM (Life technologies) containing 10% FBS
(Life technologies) at 37C in the presence of 5% CO 2 by BioStaion CT(Nikon
Corporation). All images (.bmp files) were processed using MetaMorph software
(Molecular Devices) and our original programs with original filter sets, as described
for the regression analysis model. Using integrated morphometry analysis, the
number of objects was measured together with seven individual morphological
features in MetaMorph, such as total area, breadth, fiber length, fiber breadth, shape
factor, elliptical form factor, and inner radius. Prior to statistical analysis, all
morphological data of the noise objects (non-cellular objects) were automatically
removed by the original noise-reduction algorithm (image auto-wash method). The
features were averaged in 5 view fields from the same well. For each feature,
average and average change ratio were calculated for and between each time point
(24, 48, 72, and 96 h), and a total of 203 features were used as input features. In the
modeling process, input features were selected by stepwise parameter selection in
the linear discriminant analysis using SPSS software (for Windows 11.5.1).
For the target event, we intentionally designed two conditions of trypsin treatment
during cell passage: 0.25% trypsin treatment (too dense = error) and 0.025% trypsin
Search WWH ::




Custom Search