Giovanni F. Crosta

Image analysis and automatic classification of transformed foci

Carcinogenesis is a multi‐step process involving genetic alterations and non‐genotoxic mechanisms. The in vitro cell transformation assay allows the monitoring of the neoplastic phenotype by foci formation in suitable cells (e.g. C3H10T1/2 mouse embryo fibroblasts) showing aberrant morphology of massive build‐up, polar and multi‐layered densely stained cells. The classification of transformed foci in C3H cells relies on light microscopy scoring by a trained human expert based on standard rules. This procedure is time‐consuming and prone, in some cases, to subjectivity, thereby leading to possible over‐ or under‐estimation of the carcinogenic potential of tested compounds. Herewith we describe the in vitro neoplastic transformation induced by B[a]P and CdCl2, and the development of a foci classifier based on image analysis and statistical classification. The image analysis system, which relies on ‘spectrum enhancement’, is quantitative and extracts descriptors of foci texture and structure. The statistical classification method is based on the Random Forest algorithm. We obtained a classifier trained by using experts supervision with a 20% classification error. The proposed method could serve as a basis to automate the in vitro cell transformation assay.

Estimating structural damage of the cytoskeleton by means of morphological descriptors

Direct methods (spatial differentiation), fractal analysis and spectral analysis (“spectrum enhancement”) have been applied to extract morphological descriptors from images of cytoskeletal microtubules. Images had been obtained from experiments on cultured cells (rat hepatocytes). Principal components analysis has been applied to morphological descriptors. An image classifier has thus been trained to tell normal (control) microtubule structures from those treated by a given concentration of a fungicide for a given time. Validation has been performed on sets of new images of the same two classes. Then the classifier has been used to rank the morphology of microtubules treated at lower doses and to quantify structural recovery after exposure. The paper is the first account of extensive morphological classification of microtubules and paves the way to a dose-response relationship based on quantitative morphology.

Response of cytoskeletal microtubule organization to a xenobiotic estimated from image classification

These authors have been developing for some years a variety of morphological classifiers, which analyse images, extract descriptors by FOURIER analysis, fractal analysis and spatial differentiation, fuse these descriptors by means of multivariate statistics. Classifiers have been trained, validated and applied to recognizing patterns belonging to new classes. One of the most relevant application has been the quantitative morphology of microtubule organisation. Results, which have been described in a number of publications, have consisted of: a) the quantitative assessment of structural damage caused by xenobiotics and the ensueing recovery, and b) the estimation of dose- and time- response relations. This paper, in addition to presenting a survey of the classification methods and the related results, will focus on some instructive class-wide and cell-wise statistical properties deduced from the first principal component only. These properties lead to three questions about the dose-response behaviour of microtubules which are still open.

Quantitative morphology of cytoskeletal organization: new classifier architectures and applications

Recently, these authors developed a heterogeneous, one-level image classifier (CH) based on morphological descriptors from direct domain analysis (spatial differentiation), fractal analysis and “spectrum enhancement” (a kind of non-linear filtering). Classifier CH was applied to epi-fluorescence microscope images of cytoskeletal microtubules and was trained to recognize structural alterations of the cytoskeleton in various circumstances. The application dealt with images of rat hepatocytes (rh). The scope of this paper is twofold: a) to investigate different classifier architectures, which include the multiobjective optimization of some image analysis parameters by means of suitable algorithms; b) to apply said classifiers to new sets of images obtained from mouse fibroblasts (mf) and HepG2 (hg) cells. Image sets from control and treated cell cultures are analyzed. Classifier CH is applied to mf microtubules. A new classifier entirely relying on “spectrum enhancement” (although on different descriptors) is developed and applied to rh and hg images. From the latter classifier, by bringing in descriptors from direct domain and fractal analysis, a hierarchical classifier is derived and applied to rh images. Results are compared. Classifier performance is expressed in terms of sensitivity, specificity and information contents of the first three principal components.

Colony scoring in hematotoxicity assays by an image stack classifier: first results

In order to evaluate the potential hematotoxicity of xenobiotics, including candidate anti-cancer drugs, in vitro models of hematopoiesis are used, which involve clonogenic assays on CFU-GM (Colony Forming Unit-Granulocyte-Macrophage). These assays require live and unstained colonies to be counted. Most laboratories still rely on visual scoring, which is time consuming and error prone. As a consequence automated scoring is highly desired. A classification algorithm aimed at emulating the colony recognition and scoring capabilities of a human expert has been developed. A first account will be given herewith. Assays were carried out on CFU-GM progenitors derived from human umbilical cord blood cells and grown in methylcellulose. A three-dimensional (3-D) medium is essential for these assays to simulate the clonogenetic process which takes place in bone marrow. Stacks of images representing slices of a 3-D domain were acquired. Structure and texture information was extracted from each image. Classifier training was based on a 3-D colony model applied to the image stack. The number of scored colonies (assigned class) was required to match the count supplied by the human expert (class of belonging). Successful applications to scoring colonies, which partially overlap and/or are masked by caustics, are described. Whereas the industrys scoring methods all rely on image structure alone and process 2-D data, the classifier described herewith takes texture into account and fuses 3-D dtat from a whole stack.