Inherent genetic mutations result in individual variation in the response to the same drug; therefore, establishment of personalized drug selection techniques is important for appropriate treatment decisions, with further extensions to technologies fo...
Inherent genetic mutations result in individual variation in the response to the same drug; therefore, establishment of personalized drug selection techniques is important for appropriate treatment decisions, with further extensions to technologies for implementation of precision medicine.
Since it is not possible to directly test various kinds of drugs in patients, individual reactivity to different drugs can best be assessed by collecting tissue samples from patients and conducting experiments at the cellular level.
High-throughput screening (HTS), which relies solely on cell count information with the main advantage of rapid processing, is a widely used technique for drug response determinations. Recently, high-content screening analyses have been actively conducted to determine the cell phenotype as well as to obtain cell count information based on microscopic imaging, in which image data are integrated with the data obtained through an HTS method.
Although microscopic cell imaging analysis is generally performed by well-trained experts, the results are nevertheless subjective to some extent. In addition, image data are generated for hundreds of millions of cells, resulting in a limitation of manual analysis. Therefore, image processing and multivariate features analysis are required for extracting and analyzing cell phenotype information from images.
The generation of cell images varies according to the characteristics of the cell type and imaging device, and therefore different algorithms are used for different types of images. For example, to obtain a cell phenotype profile from fluorescence microscope images or label-free microscope images of brain cancer cells, a proper pipeline should be constructed by selecting the appropriate image processing and analysis algorithms after identifying the image characteristics.
Toward this end, we here propose a method termed “Multivariate Feature Extraction and Analysis for Image-based Profiling of Cellular Phenotypes” to analyze the response of cells derived from patients with brain tumors to a target drug.
The first step involves a preprocessing method to adjust for the brightness differences among images caused by variations in the distance between the light source and the sample during image acquisition. This non-uniform illumination can be resolved through standardizing the brightness with histogram equalization and normalization to uniformly correct the intensity and improve the contrast of the images for clearer segmentation.
Second, the cell region is segmented into individual cells or nucleus units in the enhanced image using a local adaptive threshold technique with consideration of the characteristics of the image pattern in the microscope cell image. In fluorescence microscopy, the region used for determining the image brightness value and morphological characteristics of the cell object is segmented according to the culture environment.
Third, the phenotypes related to the cell object are extracted, such as the degree of fluorescence expression and morphological characteristics in the spatial and frequency domains of the segmented cell region. The phenotype is also determined based on the distribution pattern of cells in the well.
Finally, the numerical values of the above-mentioned extracted cell phenotypes are normalized, and the cell phenotype is quantitatively analyzed according to the type and concentration of drug treatment based on a statistical method and a machine-learning algorithm.
The proposed method can allow for the extraction and analysis of several cell phenotype features that differ among patients and cells in response to an anti-cancer drug. In addition, it is possible to quantitatively analyze features that are otherwise hard to detect even by the naked eye, which reduces the subjective nature of such analyses while enabling statistical analysis of the large amount of data generated.
Furthermore, through determinations of various cell parameters, the efficacy and sensitivity of the cellular response to individual cancer drugs can be quantitatively assessed. This method is expected to be applied for the appropriate selection of personalized anti-cancer drugs in precision medicine, and can further help to determine the potential biological significance of a drug according to cell phenotype.