Abstract :
Automated analysis and recognition of microscopic images of cells and tissue sections is undergoing active development. With regard to cells, the problems posed by samples of white and red blood cells presented in a monocellular layer are well under control, and automated laboratory devices are in the clinical testing stage. The analysis of white blood cell populations offers good promise in immunologic, radiation biologic, and experimental chemotherapeutic research. The situation encountered in prescreening for cervical cancer in Papanicolaou-stained gynecologic samples calls for significantly more stringent cell recognition and sample size requirements. Methods based on high-resolution digitized imagery appear currently to be the only approach with the required discriminatory power; they may evolve as practical devices, however, only when special computers with fully parallel processing capabilities are available. As applied to tissue sections, the complexity of the scene-segmentation, analysis, and classification problems has generally restricted research work to the area of feature extraction and to the determination of descriptive stereologic parameters. Processors are needed with an architecture more compatible with the two-dimensional information of pictorial data and with suitable operating systems. The field of microscopic image analysis demands competence in a wide range of disciplines, from the practical clinical situation to optics, electrooptical devices, digital logic design, computer architecture, software engineering, multivariate statistics, and decision theory. A truly interdisciplinary research team is difficult to form and maintain, but is rewarding: It is only in such close cooperation that one discovers such research needs as the fact that most of the basic applied mathematical tools for what the clinician really wants to know have not yet been developed.
