Pattern Recognition Techniques - Biomedical Signal Analysis: Contemporary Methods and Applications - page 202

monary compartments. Seventeen children (f=10; mean age 5.3 years;
range 0.3-17.3 years) had chronic bronchitis (CB) . Twenty children (f=7;
mean age 6.8 years, range 2 months - 18.8 years) had interstitial lung
diseases (ILD) . In all children a bronchoalveolar lavage was performed
for diagnostic and/or therapeutic indications. Cells were obtained from
bronchoalveolar lavage fluid (BALF), and the frequency of lymphocytes
expressing different chemokine receptors (CXCR3+, CCR5+, CCR4+,
and CCR3+) which control lymphocyte migration was analyzed by four-
color flow cytometry on CD4+ and CD8+ T cell subsets. To evaluate the
contribution of the corresponding chemokines to the local effector cell re-
cruitment, the ligands for CXCR3 and CCR5, termed IP-10 (Interferon- γ
inducible Protein of 10 kDa), and RANTES (Regulated upon Activation
Normal T cell Expressed and Secreted) were quantified in BALF with a
commercial enzyme-linked immunosorbent assay (R&D Systems, Min-
neapolis, Minnesota USA).
Signal analysis
We analyzed the following parameters in BALF (visualization in figure
6.18): RANTES relative to the cell number in BALF (RANTESZZ), IP-
10, CD4+ T cells, CD8+ T cells, the ratio of CD4+ to CD8+ T cells
(CD4/CD8), CD19+ B cells, CCR5+CD4+ cells, CXCR3+CD4+ cells,
CXCR3+CD8+ cells, macrophages (M), lymphocytes (L), neutrophile
granulocytes (NG), eosinophile granulocytes (EG), the total cell count
in BALF (ZZ), systemic corticosteroid therapy (CORTISONE), and C-
reactive protein (CRP).
Altogether, we had a data set of 30 parameters; however, some
parameters were missing for some of the patients. In the following
we will use preselected subsets of these parameters as specified in the
corresponding section.
Self-organizing maps
SOMs approximate nonlinear statistical relationships between high-di-
mensional data items by easier geometric relationships on a low-di-
mensional display. They also perform abstraction by reducing the in-
formation while preserving the most important topological and metric
relationships of the primary data. These two aspects, visualization and
abstraction, can be utilized in a number of ways in complex tasks such