Proper classification plays a vital role in early leukemia discovery and patient treatment decision, especially in Acute Myeloid Leukemia (AML) which has various subtypes because of its complicated differentiation routes. There are many AML classification methods, for example, the cytogenetic method based on cell line gene mutation type which can group AML patient into good, intermediate or poor risk categories. However, there are some problems in the method. First, a large amount of clinical observation should be made to assess patient treatment responses and risk category, which could take months or even years: the right and early opportunity to treat patient will be ignored; second, the whole experiment process strongly relies on antibodies and special chemicals, exerting a strict requirement on researchers of their experiences and abilities on bio-medical area; third, the classification system relies on case studies of AML patient’s clinical responses. Depending on different research group standards and observation parameters, there are differences in assigning cytogenetic abnormalities to prognostic groups.

Living cell is confronted to varying forces from its environment. During life processes, cell mechanics is closely related to different physiological functions. Cell mechanical properties studies helps us better understand different physiological phenomena and can supplement the cytogenetic methodology. Measuring cell mechanical properties takes less time than clinical observation and offers researchers with non-biomedical background an opportunity to study AML. This method has the potential to duplicate in large quantities, which makes quickly analyze large number of data and evaluate them in short time possible. Moreover, studying cell mechanical properties does not rely on any chemicals. Thus, chemical costs can be cut down to minimum compared to immunologic and cytogenetic methods.

This thesis proposed a cell-mechanical-properties-based method to classify different myeloid leukemia cell lines from different risk categories, and to discover the cell mechanics and leukemia risk category relationship. Cluster analysis, which can classify unlabeled cell lines by the force-curve characteristics, will be proposed to figure out problem that cell mechanics is hard to be applied on clinical tests. Confocal images are presented to analyze cytoskeleton structure characteristics and support cell mechanical properties results.

Filamentous actin fiber (F-actin) is a kind of cytoskeleton under cell membrane which serves as skeleton supporting whole cell and main contributor to cell mechanical properties. Studying its characteristic is essential to understand cell mechanical properties. In the third chapter, four AML model cell lines from good and intermediate risk categories and one CML cell line with similar clinical phenomenon as AML patient are introduced. Cell line F-actin cytoskeleton structure is imaged and studied. By treating the cell with cytoskeleton polymerization inhibitor, F-actin thickness and image gray value decrease with increasing the treating concentration. F-actin thickness and gray value are different among the five cell lines.

To select a proper treatment program for a patient, it is important to have a clear classification method. In the fourth chapter, leukemia cell lines with different risk categories are used to study cell mechanical properties by AFM. By using the depth-sensing analysis, it is shown that the cell Young’s modulus value varies with depth: from higher value at small deformation to lower value in deep part. Five different cell lines can be differentiated according to the Young’s modulus value when deformation is zero. By relating cell lines’ gene mutation type and patient case history with the Young’s modulus value, it has been proved patient lifetime from disease entry to death and risk categories can be predicted.

Physics method is easy to duplicate in large quantities; therefore, it has the potential to be applied in clinical inspections. In the fifth chapter, cluster analysis is introduced to recognize unlabeled force-indentation curves. By K-means cluster analysis iteration process with selected feature vectors, the cell force-indentation curve can be successfully grouped into five clusters with over 60% of one cell line type in each cluster. The analysis process can be realized by programs, which gives opportunity to people without the professional background to utilize the method. This process helps cluster large quantities of unknown cell force-indentation curves, find out recommended experiment setting range and realize quickly detecting abnormal curves and assess the risk category. The mechanical property-based method has the possibility to be applied on disease screening in a short time.