Automatic pattern recognition for clinical medicine

Abstract

The biomedical sciences characteristically deal with huge masses of data, much of which is in the form of pictures. In this paper illustrations are described of work already accomplished by the author in the pattern recognition of biomedical pictures. The FIDAC device, an on-line, real-time, high-speed, high-resolution flying-spot computer-input device, is utilized for putting the picture into the computer's memory. Then the computer analysis proceeds by means of two programming systems: the FIDACSYS system for basic picture handling and object recognition; and the BUGSYS language used to facilitate picture analysis and measurement. In the analysis of white blood cells, the computer has been used to distinguish monocytes by means of the shape of their nuclei, as characterized in terms of significant concavities and convexities. Automatic determination of mitotic index is facilitated by having the computer count the relative number of concavities on the boundary of each nucleus on a slide of mitotic cells. Quantitative measurement of bone mineral is accomplished by utilizing the FIDAC-computer system as a selective densitometer. Automatic chromosome analysis and karyotyping is performed by means of syntax-directed boundary analysis to identify the arm ends and centromere of each chromosome, from which measurements are made for typing and normality determination. And in neuron analysis, the computer has been used to identify the cell body and locate the junction, branching points, and termini of each dendrite.