Automated Recognition of Robotic Manipulation Failures in High-throughput Biodosimetry Tool.

A completely automated, high-throughput biodosimetry workstation has been developed by the Center for Minimally Invasive Radiation Biodosimetry at Columbia University over the past few years. To process patients' blood samples safely and reliably presents a significant challenge in the development of this biodosimetry tool. In this paper, automated failure recognition methods of robotic manipulation of capillary tubes based on a torque/force sensor are described. The characteristic features of sampled raw signals are extracted through data preprocessing. The twelve-dimensional (12D) feature space is projected onto a two-dimensional (2D) feature plane by the optimized Principal Component Analysis (PCA) and Fisher Discrimination Analysis (FDA) feature extraction functions. For the three-class manipulation failure problem in the cell harvesting module, FDA yields better separability index than that of PCA and produces well separated classes. Three classification methods, Support Vector Machine (SVM), Fisher Linear Discrimination (FLD) and Quadratic Discrimination Analysis (QDA), are employed for real-time recognition. Considering the trade-off between error rate and computation cost, SVM achieves the best overall performance.

Automatic detection of unstained viable cells in bright field images using a support vector machine with an improved training procedure.

Detection of unstained viable cells in bright field images is an inherently difficult task due to the immense variability of cell appearance. Traditionally, it has required human observers. However, in high-throughput robotic systems, an automatic procedure is essential. In this paper, we formulate viable cell detection as a supervised, binary pattern recognition problem and show that a support vector machine (SVM) with an improved training algorithm provides highly effective cell identification. In the case of cell detection, the binary classification problem generates two classes, one of which is much larger than the other. In addition, the total number of samples is extremely large. This combination represents a difficult problem for SVMs. We solved this problem with an iterative training procedure ("Compensatory Iterative Sample Selection", CISS). This procedure, which was systematically studied under various class size ratios and overlap conditions, was found to outperform several commonly used methods, primarily owing to its ability to choose the most representative samples for the decision boundary. Its speed and accuracy are sufficient for use in a practical system.

A new preprocessing approach for cell recognition.

In this paper, we describe a novel strategy for combining fisher's linear discriminant (FLD) preprocessing with a feedforward neural network to classify cultured cells in bright field images. This technique was applied to various experimental scenarios utilizing different imaging environments, and the results were compared with those for the traditional principal component analysis (PCA) preprocessing. Our FLD preprocessing was shown to be more effective than PCA due in large part to the fact that FLD maximizes the ratio of between-class to within-class scatter. The new cell recognition algorithm with FLD preprocessing improves accuracy while the speed is suitable for practical applications.

An automated imaging system for radiation biodosimetry.

We describe here an automated imaging system developed at the Center for High Throughput Minimally Invasive Radiation Biodosimetry. The imaging system is built around a fast, sensitive sCMOS camera and rapid switchable LED light source. It features complete automation of all the steps of the imaging process and contains built-in feedback loops to ensure proper operation. The imaging system is intended as a back end to the RABiT-a robotic platform for radiation biodosimetry. It is intended to automate image acquisition and analysis for four biodosimetry assays for which we have developed automated protocols: The Cytokinesis Blocked Micronucleus assay, the γ-H2AX assay, the Dicentric assay (using PNA or FISH probes) and the RABiT-BAND assay.

Adapting the γ-H2AX assay for automated processing in human lymphocytes. 1. Technological aspects.

The immunofluorescence-based detection of γ-H2AX is a reliable and sensitive method for quantitatively measuring DNA double-strand breaks (DSBs) in irradiated samples. Since H2AX phosphorylation is highly linear with radiation dose, this well-established biomarker is in current use in radiation biodosimetry. At the Center for High-Throughput Minimally Invasive Radiation Biodosimetry, we have developed a fully automated high-throughput system, the RABIT (Rapid Automated Biodosimetry Tool), that can be used to measure γ-H2AX yields from fingerstick-derived samples of blood. The RABIT workstation has been designed to fully automate the γ-H2AX immunocytochemical protocol, from the isolation of human blood lymphocytes in heparin-coated PVC capillaries to the immunolabeling of γ-H2AX protein and image acquisition to determine fluorescence yield. High throughput is achieved through the use of purpose-built robotics, lymphocyte handling in 96-well filter-bottomed plates, and high-speed imaging. The goal of the present study was to optimize and validate the performance of the RABIT system for the reproducible and quantitative detection of γ-H2AX total fluorescence in lymphocytes in a multiwell format. Validation of our biodosimetry platform was achieved by the linear detection of a dose-dependent increase in γ-H2AX fluorescence in peripheral blood samples irradiated ex vivo with γ rays over the range 0 to 8 Gy. This study demonstrates for the first time the optimization and use of our robotically based biodosimetry workstation to successfully quantify γ-H2AX total fluorescence in irradiated peripheral lymphocytes.

Multiclass detection of cells in multicontrast composite images.

In this paper, we describe a framework for multiclass cell detection in composite images consisting of images obtained with three different contrast methods for transmitted light illumination (referred to as multicontrast composite images). Compared to previous multiclass cell detection results [1], the use of multicontrast composite images was found to improve the detection accuracy by introducing more discriminatory information into the system. Preprocessing multicontrast composite images with Kernel PCA was found to be superior to traditional linear PCA preprocessing, especially in difficult classification scenarios where high-order nonlinear correlations are expected to be important. Systematic study of our approach under different overlap conditions suggests that it possesses sufficient speed and accuracy for use in some practical systems.

The RABIT: a rapid automated biodosimetry tool for radiological triage.

In response to the recognized need for high throughput biodosimetry methods for use after large-scale radiological events, a logical approach is complete automation of standard biodosimetric assays that are currently performed manually. The authors describe progress to date on the RABIT (Rapid Automated BIodosimetry Tool), designed to score micronuclei or gamma-H2AX fluorescence in lymphocytes derived from a single drop of blood from a fingerstick. The RABIT system is designed to be completely automated, from the input of the capillary blood sample into the machine to the output of a dose estimate. Improvements in throughput are achieved through use of a single drop of blood, optimization of the biological protocols for in situ analysis in multi-well plates, implementation of robotic-plate and liquid handling, and new developments in high-speed imaging. Automating well-established bioassays represents a promising approach to high-throughput radiation biodosimetry, both because high throughputs can be achieved, but also because the time to deployment is potentially much shorter than for a new biological assay. Here the authors describe the development of each of the individual modules of the RABIT system and show preliminary data from key modules. System integration is ongoing, followed by calibration and validation.

DESIGN AND PRELIMINARY VALIDATION OF A RAPID AUTOMATED BIODOSIMETRY TOOL FOR HIGH THROUGPUT RADIOLOGICAL TRIAGE.

This paper presents design, hardware, software, and parameter optimization for a novel robotic automation system. RABiT is a Rapid Automated Biodosimetry Tool for high throughput radiological triage. The design considerations guiding the hardware and software architecture are presented with focus on methods of communication, ease of implementation, and need for real-time control versus soft time control cycles. The design and parameter determination for a non-contact PVC capillary laser cutting system is presented. A novel approach for lymphocyte concentration estimation based on computer vision is reported. Experimental evaluations of the system components validate the success of our prototype system in achieving a throughput of 6,000 samples in a period of 18 hours.