Image-Based Cell Quality Assessment: Modeling of Cell Morphology and Quality for Clinical Cell Therapy - Computational Modeling in Tissue Engineering - page 209

Fig. 6 Prediction results
from the human error
detection model. Each plot
represents 1 image dataset.
Morphological features in the
model were selected by a
stepwise parameter selection
process during model
construction

4

SAFE

3

ERROR

2

1

0

-1

-2

-3

-4

which relates to cells in clinical cell therapy, is often very complicated and
ambiguous. Machine learning researchers should be aware that most cellular
qualities are too complex to define using a single value. For example, although
various biomarkers to define ''stemness'' of cells are measured, there exist various
profiles because not all the markers are completely the same in cells. Therefore,
even with the detailed profiling of biomarkers with flow cytometry analysis, cell
qualities are commonly discussed as ''levels'' or ''categories.'' There are many
more examples of such categorical definitions of cell qualities, such as cell lineage
or cell damage rate. For such cellular qualities, there is still no good single marker
to define the phenomenon quantitatively.
Among such complex quality conditions of cells, ''cell-type classification'' is a
strongly required solution in clinical cell therapy. In clinical cell therapy, a cell
source is obtained from the tissues of patients. Cells are isolated from tissues by a
process of primary culture, including enzyme digestion and explant culture process.
Because a tissue is a complex of different cell types, there is a high possibility of
incorporating ''unwanted cells,'' which are designated as ''cell contamination,'' in
the primary cell population. For example, in the dermal defect therapy using fibro-
blasts, keratinocyte contamination increases the risk of dermal cyst formation.
Similarly, in skin burn treatment, fibroblast contamination in keratinocytes causes a
risk of skin contraction. Although such cell types are clearly known, there are still
cells without good markers, such as ''fibroblasts.'' Without a clear marker protein,
fibroblasts cannot be stained or detected in the keratinocyte population. Furthermore,
if clinical doctors require detection of keratinocytes among fibroblasts, staining of
cells for keratinocyte-specific protein markers depletes precious patients' cell
source. Therefore, ''cell type classification'' is an important challenge for image