The hardware and software components are described to track many flagellates simultaneously in real time. The technique is based on on-line digitization of four frames taken at 80 ms intervals and stored in a specialized video memory. The outline and position of each organism are determined by chain coding and followed through the recorded series of images. The movement vectors of all organisms tracked are stored in the form of deviation angles from a predefined stimulus (light or gravity) direction and the distance each organism has moved in the time interval determined by the hardware clock of the computer. Subsequent programs allow one to determine circular histograms of movement directions and velocities in dependence of the movement direction. Examples of several orientation patterns are given for both photoorientation and gravitaxis in the dinoflagellate, Peridinium gatunense.

Simultaneous tracking of flagellates in real time by image analysis

Method and apparatus for analyzing a 2-dimensional sensory-input field containing three or more distinct classes of objects. The field is represented by a 2-dimensional array of elements in which each separate class of objects in the field has a separate classification value. The field is scanned with a 2×2 window containing the element being scanned and three adjacent and connected window elements, where each of the adjacent window elements has been previously assigned an object label which identifies one object containing that element. If the element being scanned has the same classification value as any of the adjacent window element, it is assigned the object label of that same-classification window element. Otherwise, the element being scanned is assigned a new label. If the element being scanned has the same classification value as two adjacent and orthogonal elements, and these orthogonal elements have been previously given different object labels, the two different object labels are merged. The result is to identify each separate object in the image. Geometric properties of the objects can be calculated during the scanning procedure.

Object analysis of multi-valued images

In many departments of cytology, cytogenetics, hematology, and pathology, research projects using high-resolution computerized microscopy are now being mounted for computation of morphometric measurements on various structural components, as well as for determination of cellular DNA content. The majority of these measurements are made in a partially automated, computer-assisted mode, wherein there is strong interaction between the user and the computerized microscope. At the same time, full automation has been accomplished for both sample preparation and sample examination for clinical determination of the white blood cell differential count. At the time of writing, approximately 1,000 robot differential counting microscopes are in the field, analyzing images of human white blood cells, red blood cells, and platelets at the overall rate of about 100,000 slides per day. This mammoth through-put represents a major accomplishment in the application of machine vision to automated microscopy for hematology. In other areas of automated high-resolution microscopy, such as cytology and cytogenetics, no commercial instruments are available (although a few metaphase-finding machines are available and other new machines have been announced during the past year). This is a disappointing product, considering the nearly half century of research effort in these areas. This paper provides examples of the state of the art in automation of cell analysis for blood smears, cervical smears, and chromosome preparations. Also treated are new developments in multi-resolution automated microscopy, where images are now being generated and analyzed by a single machine over a range of 64:1 magnification and from 10,000 X 20,000 to 500 X 500 in total picture elements (pixels). Examples of images of human lymph node and liver tissue are presented. Semi-automated systems are not treated, although there is mention of recent research in the automation of tissue analysis.

High-resolution image analysis.

Automated computer-controlled image analysis of digitized video images has developed into a versatile tool useful for many applications. Both on the macroscopic and microscopic scales counting, identification and tracking are facilitated by these techniques which allow a quantitative, fast and objective analysis. Though image analysis has not yet reached the capabilities of the human eye and brain, it helps in both routine analysis and research tasks. Future enhancements made possible by the availability of larger memories and higher processing speeds will enhance the realm of applications. Especially the use of pipeline processing and dedicated hardware for the routines of image manipulation, including averaging and Laplace filtering, will open new fields of use. The most difficult task to master will be the extraction of complex parameters necessary for object recognition in an image, which requires a high degree of built-in intelligence both on the hardware and software level. Another ambitious task to tackle is the analysis of image sequences in order to track several individuals in a population in parallel.

Computer-assisted image analysis in biological sciences

After four decades of research, automation of certain aspects of the visual inspection of both cells and tissue has become a reality. This chapter not only reviews the current status of the automation of image analysis in cytology and histology, but also presents a historical account of this field extending from its inception.

Automated image processing for cells and tissue

This paper describes a machine vision system for automatic characterization of sections of biopsied tissue. Color images are first segmented by using an original color segmentation procedure. A structural analysis is then performed over the segmented images to extract the same features that a pathologist would use to make a diagnosis. Alternatively, monochromatic images can be analyzed by using cellular logic techniques. These two methods were used to implement a machine vision system capable of recognizing normal liver biopsies and liver biopsies showing alcoholic hepatitis and acute viral hepatitis. The color segmentation procedure leads to 100 percent classification success in distinguishing between the normal and acute viral hepatitis classes. By using cellular logic techniques, no overlap was found between the normal, acute viral hepatitis, and alcoholic hepatitis classes at the 90 percent confidence interval.

David Yorkgitis

Papers

A Study of Multidimensional Multicolor Images