Workpackage Leader
Partners Objectives Description of work Work in progress Deliverables and Milestones Achievements Partners
Due to the scarcity of foetal cells in the maternal circulation, even the best current enrichment strategies result in preparations where in the overwhelming majority of cells are of maternal origin and which contains only a few foetal cells, i.e. about 1 in ten thousand instead of 1 in ten million in un-enriched maternal blood. It is, furthermore, also very likely that a significant number of foetal cells are lost during the enrichment procedure. For this reason we will exploit recent developments in automated image analysis, scanning, identification and laser-assisted micro-manipulation of rare events in order to address these pitfalls. In order to achieve this, an Algorithms Resource Centre (ARC) will be established in Oxford, which will liase with research groups engaged in the development and examination of new foetal cell specific markers (WP1) and leading European SME's highly proficient in rare cell detection and manipulation. The ARC/SME liaison will focus on image processing, pattern classification, applicability of a knowledge based approach and requirement of the user interface. This larger consortium will aid in determining the parameters required for optimal rare event analysis e.g. nuclear (FISH or telomeres) or cellular markers (mRNA or protein), fluorescent characteristics or bright-field conditions. These will then be tested on enriched samples as well as on maternal whole blood preparations. Foetal cells identified by this manner can then re-examined used cutting edge genomic /proteomic or nano-technologies for the further characterisation of highly specific foetal cell markers (WP1.4 to 1.6) WP 2.1 Provision of Algorithms Resource Centre
WP2.2 Improved Statistical Model for MLPA Analysis
WP2.3 Detection, relocation, and recognition of fetal cells.
WP2.1 Provision of Algorithms Resource Centre
WP2.2 Improved Statistical Model for MLPA Analysis
WP2.3 Detection, relocation, and recognition of fetal cells
These deliverables and milestones are from JPA M49 to 60 (1 March 08 to 28 Feb 09). WP2 Deliverables
WP2 Milestones
Current experimental procedures result in images that are of sufficient quality for an expert to interpret, but which have too much unpredictable variation for reliable processing by algorithm. Therefore, we conducted a 360 degree study, involving selected SAFE partners, to determine the nature and extent of experimental variation. The results of this study:
The development of reliable image interpretation algorithms is a very difficult mathematical problem that has challenged investigators for over 30 years. More progress is necessary to advance the state of the art. Therefore, we have produced the most comprehensive survey undertaken so far on computer vision techniques for cytometric segmentation. We have also developed new algorithms for highly accurate segmentation of nuclei and detection of fluorescent markers. Previous laboratory practice has underestimated the amount of computing power necessary to make accurate interpretations of nucleic markers and to process databases containing thousands of images. Therefore, we have commissioned a 32-processor HPC (High Performance Computer) cluster server available via the Internet to the SAFE
To exploit the experience of the SAFE consortium, all partners were canvassed about their usage of automated fluorescence microscopy. Eight partners use their equipment daily for routine experiments, six citing FISH interphase analysis as their most common requirement. FISH spot counting is the most common quantitative measurement.
Telomere mass is substantially greater in foetal cells than maternal cells as it decreases in neonates at a rate of about 3 kbp/yr decreasing to about 50 base pairs per year from the age of 4. Therefore Oxford Brookes University (Partner 4) and the University of Warwick (SAFE Partner 1a) engaged in collaborative research to investigate telomere mass as a general discriminator of foetal cells. The experimental objective was therefore to determine how distinguishable are foetal and maternal telomeres using FISH labelling and conventional wide field fluorescence microscopy. The University of Warwick produced a database of 2,166 images with approximately 14,000 human foetal and adult leucocyte nuclei and 315,000 probed telomere regions for this purpose. The telomere specific fluorescent marker probes appear as bright spots in the microscope image, the greater the mass, the brighter. To quantify the emission associated with telomere mass it is required to a) identify usable cells, b) identify all spots in the nuclei and c) for each nucleus to measure the total fluorescent emission from the probed material. Due to the large quantity of data to be evaluated, automation of the above processes was required for the results to be produced in a reasonable time. An important component was the development of appropriate computer vision techniques for automated analysis of fluorescence microscopy images, which was the remit of Oxford Brookes University. Because the cells are uncultured, their nuclei have widely varying size, shape and texture. Therefore a novel histogram based segmentation algorithm was developed that correctly segments over 99% of cases, compared with 80% for the commonly used watershed algorithm regarded as the state of the art. A new algorithm was also developed
Experimental results show that the foetal material tends to have brighter emission than adult, but measurements have a high variance and there is considerable overlap between the parental and foetal histograms. As with any physical quantity, telomere measurements contain errors, or noise, of various kinds. It was therefore decided to obtain an improved understanding of the characteristics of the imaging system and the detected signals. An algebraic description was developed of the relationships between the histogram of a captured image, its underlying signal and the perturbing effects of noise. Dependent, independent, additive and multiplicative noise was considered as well as a general functional relationship. All of the algebraic results could be expressed as convolutions of various forms. The expressions also have inverses, giving the opportunity to de-convolve the noise-free statistics from noisy data, if the form of the noise and its functional relationship to the data is known. To give a further improved model of a system for quantification of telomere content, the image acquisition characteristics were formally examined with respect to Shannon channel capacity. It was shown that the channel capacity increases much more rapidly with the number of sensing elements than that of the signal quantisation intervals. As a design principle, it is therefore more efficient to increase the spatial resolution of the sensor than the number of quantisation levels. Medical University Graz (Partner 8b) have isolated trophoblast cells from interruption material and labelled them according to their slide-based staining technique by means of immunofluorescence using different antibodies directed against the trophoblast cell surface. After scanning different slides, a classifier was created using the MetaSystems RCDetect software. At Athens University (Partner 19) approximately 100 NRBCs isolated with the above method were captured and stored using MetaSystem's instrument. These pictures are available for the determination of various parameters which could allow the differentiation between foetal and maternal cells. P.A.L.M. (Partner 47) have recently developed an automated protocol for software-driven isolation and lift-up for non-contact Laser Capture Microdissection directly on-chip on different single cell samples. They have also demonstrated PCR on single buccal cells and single blood cells on-chip. When MLPA was initially introduced, the results were analysed by visually comparing the probe signals with those from the respective control experiments. More recently, statistical analyses have been carried out on the numerical data, generating quantitative results. Various software packages for automated or semi-automated MLPA analysis are now appearing. Existing methods of MLPA analysis were reviewed and proposals made for improving MLPA data processing. A general-purpose MLPA analysis tool has therefore been designed to investigate alternative data processing models. |
