Biomedical Engineering Reference
In-Depth Information
laboratory experts. The automation of AAB IIF reading, including pattern rec-
ognition, would therefore be effective in reducing intra- and inter-laboratory
variability, and meeting the growing demand for cost-effective assessment of large
numbers of samples (i.e., high throughput). This automated IIF interpretation
system has previously been developed for ANA (anti-nuclear antibody) detection
on HEp-2 cells, assessment of dsDNAab on Crithidia luciliae and of ANCA
on human neutrophiles and found to be reliable for the positive/negative differ-
entiation of IIF patterns as well as pattern recognition of ANA and ANCA findings
[
29
,
30
,
35
]. The AKLIDES
system uses common standard protocols for IIF
staining of immobilized cells on glass slides by the AAB to be detected. A sec-
ondary antibody labeled with FITC is employed to reveal the specific AAB
binding. Additionally, 4
0
,6-diamidino-2-phenylindol (DAPI) staining is used for
focusing procedures and nucleus detection. Image acquisition, evaluation of image
quality, image segmentation for object detection, feature extraction for object
description, and classification of objects have been automated to provide a stan-
dardized detection method for even subcellular structures. Images of IIF patterns
are automatically taken and autofocusing is the most critical step in IIF reading
since insufficient focusing in the range of a few microns may bring about false
classification of staining patterns [
36
]. Typical IIF images obtained by AKLIDES
are shown in Fig.
10
.
Currently, cell-based IIF is usually interpreted by an experienced immunologist
in routine laboratories, assessing areas of the image with sufficient quality for their
pattern-recognition decision. In terms of automated reading, certain image areas
are not suitable for interpretation due to artifacts or staining failures. This creates
the need to implement a preselection algorithm for suitable scenes for pattern
analysis. Taking into account the structure and the staining of adherent cell
monolayers, quality criteria, e.g., the maximum plausible cell size, were defined
for proper image acquisition. For quality control of image reading, the quality of
images is measured by descriptive parameters (e.g., sharpness, brightness, number
of cells). Each image below a certain quality level is excluded from further pattern
analysis.
This guarantees that only correctly acquired scenes are processed by the fol-
lowing more complex algorithms. To differentiate distinct patterns, IIF image
signals were processed further in hierarchical processing steps: (i) positivity, (ii)
localization (nucleus, cytoplasm, chromatin of mitotic cells), and (iii) main nuclear
pattern. For the deep analysis of acquired image data, the following steps were
processed sequentially: acquisition, quality control, segmenting, description of
object and object classification. Segmented objects were defined finally by
boundary, regional, topological, and texture/surface descriptors. Digital features
were combined into rules, resembling the rules employed by experts for pattern
recognition.
In summary, accumulating data have shown a good agreement between
AKLIDES
and manual IIF interpretation for ANA, ANCA, and dsDNA antibody
detection in patients with autoimmune diseases. We therefore conclude that
