G. S. Lodwick

A Survey of Preprocessing and Feature Extraction Techniques for Radiographic Images

Feature extraction is one of the more difficult steps in image pattern recognition. Some sources of difficulty are the presence of irrelevant information and the relativity of a feature set to a particular application. Several preprocessing techniques for enhancing selected features and removing irrelevant data are described and compared. The techniques include gray level distribution linearization, digital spatial filtering, contrast enhancement, and image subtraction. Also, several feature extraction techniques are illustrated. The techniques are divided into spatial and Fourier domain operations. The spatial domain operations of directional signatures and contour tracing are first described. Then, the Fourier domain techniques of frequency signatures and template matching are illustrated. Finally, a practical image pattern recognition problem is solved using some of the described techniques.

Computer analysis of chest radiographs

A method is described for the automatic recognition of congenital heart abnormalities via digitized images of posteroanterior (PA) view chest radiographs. a closed curve representing the hearts outline is determined by computer; measurements on the outline are extracted via the Fourier descriptor technique; and these measurements are used for normal-abnormal grouping via a modified maximum likelihood classification algorithm. The system has been implemented on a dec pdp-11/20 and is capable of performing real-time diagnoses at the rate of about two minutes per radiograph.

Computer diagnosis of radiographic images

The potential of optical scanning equipment and digital computers for assisting or replacing human judgment in medical diagnosis has been recognized by investigators for some time. A number of efforts have been made, with varying degrees of success, in developing automatic techniques for recognizing and classifying blood cells, chromosome analysis and karyotyping, identifying leucocytes, and processing scintigram images obtained in nuclear medicine. These image analysis techniques are now being extended to the clinical specialty of diagnostic radiology where there is an urgent need to provide assistance in handling the several million radiographs read by radiologists each year. The need for computer-aided diagnosis in radiology is becoming increasingly urgent because of the expanding population and the continuing demand for improved quality of medical care. The use of computers in radiology can free the diagnostic radiologist from routine tasks while providing more accurate measurements that lead to consistent and reliable diagnoses.

Extraction of Connected Edges from Knee Radiographs

A computer algorithm is described that extracts the edge information from a digital image of an AP radiograph of the knee. Automated image field partitioning is used as a simplifying first step in the process. Methods that employ a priori knowledge are used to insure simple connected edges.

Digital techniques for image enhancement of radiographs

Abstract This paper deals with digital, position variant and invariant linear and nonlinear techniques for image enhancement (Rozenfeld, 1969). The radiographic system, the image processing and display system, and the human visual system are discussed. The enhancement techniques tailor image content of radiographs to the type of information most useful to the human visual system.

The computer analysis of chest radiographs

This paper describes a program for automatically extracting lung and heart features in the digitized image of posteroanterior (PA) view chest radiographs. A graph-directed analysis is used to guide the search for objects from the largest to the smallest in the radiograph. Global information is used to guide the analysis of the program. Consequently, only the points in a small range are searched and tested against local criteria to detect boundary points. The entire lung boundary is broken into four segments: upper inside boundary, lower inside boundary, boundary along the diaphragm and outside boundary. Slightly different global-local criteria for detecting the edge points along each segment have been developed and tested on 423 PA chest radiographs of patients of all ages. The results obtained indicate the program can locate the accurate boundary on all cases except infants. Twenty-seven measurements which describe the shape and size of the heart are extracted; these measurements are used for normal abnormal classification via a modified maximum likelihood classification algorithm.

MARS: Missouri Automated Radiology System

The primary role of the radiologist is to examine patients, usually with the help of ionizing radiation, in order to provide information of use in patient care. The radiologist functions as a consultant that is, patients are referred to him by many other physicians, and he delivers information obtained from his special methods of examination back to each patients referring physician. As a result the radiologist deals with greater numbers of patients than most physicians. Also, he needs to move a great deal of data quickly and accurately.

Computer Analysis Of Radiographic Images

The significant steps in computer anal-ysis of radiographic images are 1) digitization of the x-ray image; 2) preprocessing of digital images; 3) extraction of significant features; and 4) automatic classification for normal, abnormal, and differential diagnosis. A typical digital image processing facility is described, including the needed interactive type digital displays. Techniques used for preprocessing radio-graphic images are detailed; typically, these are used to ensure a higher degree of success in the later stages of digital processing. Contour tracing and region enumeration algorithms are detailed for use in the com-puter analysis of radiographs. The important descriptive approach to the problem of feature extraction is provided along with illustrative examples. A case study of rheumatic and congenital heart disease is presented for the cardiac shape analysis of PA chest films.

Learning Texture Information from Singular Photographs and its Application in Digital Image Classification

An area which has rapidly gained interest in the last few years is that of remote sensing by imagery.1, 2 Certainly the successful mission of the ERTS has contributed to an expanding interest and increased potential.3 The potential uses of such remotely sensed data cover a variety of applications, including land use studies, crop quality, accurate map making, and evaluating natural resources. In utilizing this data in a particular application, a significant problem is the amount of data that must be processed to obtain specific information or features pertinent to the application. This is inherent in image or visual information; when this is multiplied by a large number of the available images, the problem of obtaining details or small features from such images is significant. This is particularly true when some set of features is to be used in a parametric pattern recognition scheme. However, there are many instances when one is interested in identifying an area or region which may encompass many details, but is small compared with an overall image and represents a defined entity. Then, if one is interested in identifying or classifying such a region, its gross characteristics must be represented while ignoring small details that may vary significantly from one sample observation to another.

Control of discrimination level and recognition error in sequential pattern classification

This paper introduces a measure of the ability for a pattern classifier to discriminate among different pattern classes. The use of this measure then defines a compromise between pattern discrimination and recognition error. The measure is particularly useful in sequential pattern classification in allowing control of discrimination level versus recognition error. The trade-off between discrimination level and recognition error is demonstrated for data used in a medical diagnosis problem.