Geovanni Martinez

Optical Inline Measurement Procedures for Counting and Sizing Cells in Bioprocess Technology

To observe and control cultivation processes, optical sensors are used increasingly. Important parameters for controlling such processes are cell count, cell size distribution, and the morphology of cells. Among turbidity measurement methods, imaging procedures are applied for determining these process parameters. A disadvantage of most previously developed imaging procedures is that they are only available offline which requires sampling. On the other hand, available imaging inline probes can so far only deliver a limited number of process parameters. This chapter presents new optical procedures for the inline determination of cell count, cell size distribution, and other parameters. In particular, by in situ microscopy an imaging procedure will be described which allows the determination of direct and nondirect cell parameters in real time without sampling.

Cell cluster segmentation based on global and local thresholding for in-situ microscopy

This paper describes a new cell cluster segmentation algorithm based on global and local thresholding for in-situ microscopy. The global threshold is estimated by applying a known maximum likelihood thresholding technique. Assuming that the background pixels around a cluster have similar intensity values, the local threshold used to improve the segmented region after global thresholding is estimated as the average of the intensity values of a set of selected surrounding background pixels of that region. First, all pixels on the border of the segmented region are defined as possible candidates of surrounding background pixels. Then, an algorithm based on RANSAC (random sample consensus) is applied to detect outliers within the candidates. Only the inliers are used for estimation of the local threshold value. The algorithm was applied to real intensity images captured by an in-situ microscope. The experimental results show that the segmentation accuracy improved by 82%

Cell density estimation from a still image for in-situ microscopy

The paper investigates an algorithm to estimate the cell density (cell count) from a still intensity image captured by an in-situ microscope directly from inside a bioreactor. In comparison with other algorithms, ours has the advantage that it allows a reliable cell density estimation, even though the cells build clusters in the scene. First, image regions containing at least one cell are segmented by applying a maximum-likelihood thresholding technique. Then, the cell density inside each segmented region is estimated by maximizing the variance of the circular Hough transform of the edges inside it. The edges are extracted by applying the smallest univalue segment assimilating nucleus algorithm (SUSAN). The total cell density is the sum of the cell densities estimated inside the segmented regions. The proposed algorithm has been implemented and applied to thousands of real images of cultures of mammalian baby hamster kidney cells (BHK cells) captured by an in-situ microscope. The average of the percentage of the absolute cell density estimation error was 6.27%. The estimates are similar to those obtained with current off-the-shelf cell density monitoring instruments for cultures up to cell densities of 5/spl times/10/sup 6/ cells/ml.

Teleoperating ROBONAUT: A case study.

In this paper, we present a non-intrusive method for human motion estimation from a monocular video camera for the teleoperation of ROBONAUT (ROBOtic astroNAUT). ROBONAUT is an anthropomorphic robot developed at NASA - JSC, which is capable of dextrous, human- like maneuvers to handlecommonextravehicularactivity tools. The humanoperatoris represented using an articulated three-dimensional model consisting of rigid links connected by spherical joints. The shape of a link is described by a triangular mesh and its motion by six parameters: one three-dimensional translation vector and three rotation angles. The motion parameters of the links are estimated by maximizing the conditional probability of the frame-to-frame intensity differences at observation points. The algorithm was applied to real test sequences of a moving arm with very encouraging results. Specifically, the mean error for the derived wrist position (using the estimated motion parameters) was 0.57 0.31 cm. The motion estimates were used to remotely command a robonaut simulation developed at NASA - JSC.

Region and contour based cell cluster segmentation algorithm for in-situ microscopy

In this contribution a new algorithm is proposed for segmenting the image regions of the cell clusters present in a static image captured by an in-situ microscope inside of a bioreactor. A cell cluster is a group of one or more cells that are very close to each other, almost overlapping. The new algorithm combines a contour based segmentation approach with a region based segmentation approach. First, seeds are selected only in the background. To this end, image contours and the first and second moments of the pixelspsila intensity values in the background and in the cell clusters are evaluated. The moments are estimated from the histogram of the pixelspsila intensity values by applying a Maximum-Likelihood estimator. Following, the background region is extracted by region growing from the selected seeds. Finally, the segmented regions of the cell clusters are those image regions which do not belong to the previously extracted background region. Experimental results show an improvement of 33.33% in the reliability and an improvement of 55.1% in the accuracy of the cell cluster segmentation results.

Foreground segmentation of human insulin crystal images for in-situ microscopy

In this paper, an algorithm is introduced for segmenting the foreground regions present in a human insulin crystal intensity image captured by an in-situ microscope inside of a bioreactor. The segmentation is carried out by classifying all image pixels into pixels belonging to the foreground regions and pixels belonging to the background region. For classification, the local intensity variance at each pixel position is compared to a threshold. Those pixels whose local intensity variance is bigger than the threshold are classified as belonging to the foreground regions. The threshold is estimated as a linear combination of two statistical characteristics of the local intensity variance values at the pixels in the background region. Those statistical characteristics are estimated from the histogram of the local intensity variance values of all image pixels by maximizing a likelihood function using an Expectation and Maximization approach. Misclassifications are corrected by particle filtering. Experimental results on real data revealed a processing time of 11.82 seconds/image, an excellent reliability and a segmentation error of approximately 14 pixels.

Three-dimensional cell counting for in-situ microscopy

In this contribution, an algorithm is presented for counting cells in the first-, second- and third-layers of three-dimensional three-layer cell clusters from a static intensity image captured by an in-situ microscope. The number of cells in an arbitrary cell cluster is estimated as the ratio between the sum of the areas of the image regions of the parallel projections of the first-, second- and third-layers of the cluster into the image plane and the image area of a circle with a radius equal to the average cell radius of the cluster. The experimental results revealed that counting the cells also in the second- and third-layers of the clusters improved the cell count in 56% in high cell concentrations.

Intensity-Difference Based Monocular Visual Odometry for Planetary Rovers

A monocular visual odometry algorithm is presented that is able to estimate the rover’s 3D motion by maximizing the conditional probability of the intensity differences between two consecutive images, which were captured by a monocular video camera before and after the rover’s motion. The camera is supposed to be rigidly attached to the rover. The intensity differences are measured at observation points only that are points with high linear intensity gradients. It represents an alternative to traditionally stereo visual odometry algorithms, where the rover’s 3D motion is estimated by maximizing the conditional probability of the 3D correspondences between two sets of 3D feature point positions, which were obtained from two consecutive stereo image pairs that were captured by a stereo video camera before and after the rover’s motion. Experimental results with synthetic and real image sequences revealed highly accurate and reliable estimates, respectively. Additionally, it seems to be an excellent candidate for mobile robot missions where space, weight and power supply are really very limited.

Bubble segmentation based on Shape From Shading for in-situ microscopy

This paper describes a new bubble segmentation algorithm based on shape from shading for in-situ microscopy. An in-situ microscope is an instrument to capture and analyze intensity images of cells inside of a bioreactor with minimal operator intervention and without the risk of culture contamination. For bubble segmentation, the closed bubble boundaries are first extracted by thresholding a depth map. The depth map is estimated by applying the Bichsel and Pentlands Shape From Shading algorithm. Then, each extracted closed bubble boundary is filled in to obtained the bubble regions. The experimental results revealed an average processing time of 2.68 seconds and very promising bubble segmentation results.

Three-Dimensional Shape Estimation of BHK Cell Clusters from a Still Image Based on Shape from Shading for In-SITU Microscopy

In this contribution, the Lee and Rosenfelds local shape from shading (SFS) algorithm, the Tsai and Shahs linear SFS algorithm and the Bichsel and Pentlands propagation SFS algorithm are investigated with the aim of selecting the most suitable for three-dimensional shape estimation of clusters of mammalian baby hamster kidney cells (BHK cells) from an intensity image captured by an in-situ microscope in an industrial mammalian cell culture process. All three were implemented and tested using several thousand intensity images captured under varying experimental conditions. The Bichsel and Pentlands SFS algorithm was finally selected as the most suitable algorithm for three-dimensional shape estimation of BHK cell clusters. It is fast and provides less noise and more detailed depth estimates and therefore the best overall performance