Richard P. Kruger

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 Diagnosis of Pneumoconiosis

The advent of increased government involvement in occupational health maintenance with compensation for affected individuals is requiring new approaches to medical decision making. One aspect of this involvement will perhaps include the automatic mass diagnostic screening of medical films for the detection of a specffic abnormality with an occupational etiology. The results of two complementary approaches for performing diagnostic screening for presence and profusion of coal workers pneumoconiosis from the routine posterior-anterior chest radiograph are presented. The first is a digital approach utilizing the measurement of image texture, while the second uses hybrid optical-digital methods involving the optical Fourier tramsform. Both approaches yielded classification results comparable to experienced radiologists.

An Optical-Digital System For Automatic Processing Of Chest X-Rays

A description of a prototype optical-digital system for automatic processing of chest X-rays and preliminary results are presented. The system consists of a digital image scanner used for locating the lung fields and texture measurements, a diffraction pattern sampling unit which provides both annular ring and wedge samples of the Fraunhofer diffraction pattern of lung areas, and a microcomputer for control and computation of decision functions. The film is positioned by a computer-controlled film transport to provide an overall automated system. The system is currently being designed for the Appalachian Laboratory for Occupational Respiratory Diseases and will be used for screening chest X-rays of coal miners for pneumoconiosis.

Comments on "Nonstationary Assumptions for Gaussian Models in Images"

Evidence is reviewed suggesting that a Laplacian probability density function constitutes a more valid model for high-pass filtered imagery than the Gaussian model postulated in earlier work. In addition, it is shown that this discrepancy does not either seriously weaken the applicability of this class of images to a major image restoration method or challenge any other basic conclusions of the previous work.

Hybrid optical-digital radiography-based system for lung disease detection

A hybrid optical-digital system for automated measurement and disease classification of chest radiographs has been constructed. Initial results from the use of this system for classification of opacity textural patterns associated with a specific diffuse lung disease (coal workers pneumoconiosis) are presented. This system consists of three subsystems which include a digital image scanner for lung field detection, a computercontrolled X - Y transport for film positioning, and a Fraunhofer diffraction pattern sampling unit for opacity detection and measurement. A minicomputer is used to integrate these subsystems as well as render a film quality and disease classification through the use of discriminant functions and probability estimates.

A Terminal/Time Share Based Method for Interactive Left Ventricular Volume Estimation

A terminal/time share system for operator interactive left ventricular volume estimation is described. The system concept could provide a cost effective alternative to comparable minicomputer based methods and a potential for software resource sharing among several geographically remote users.

Measurements Of Lung Vascularity For Early Diagnosis Of Pulmonary Edema

Cardiogenic pulmonary edema is a major cause of pulmonary edema. With methods presently available, pulmonary edema can only be detected clinically, late in its course, making prompt therapy of the developing condition virtually impossible. The detection of an early rise in pulmonary venous pressure could allow institution of therapy and possibly prevent pulmonary edema and its accompanying morbidity and mortality. It has been shown experimentally that as pulmonary venous pressures rise, there is a redistribution of blood flow in the lungs. A prototype optical-digital system for automatic processing of coal workers pneumoconiosis chest radiographs has been developed. With the optical system it is well known that high frequency information pertains to the amount of sharpness of edge information in an image. From experience with the prototype system, it was hypothesized that lung regions with more vascularity would generate more of this high frequency than a region of lung with few vessels, and that pulmonary vascular patterns could be extracted using the Fraunhofer diffraction pattern sampling unit. To examine the effect of the number of vessels in annular ring and wedge signatures, a study was made of straight line patterns which might simulate different categories of redistribution. Different patterns were characterized by plotting the normalized energies of the annular rings and wedges. An interactive, non statistical prediction was made on a test set of line patterns with an accuracy rate of 95%. Spatial frequency signatures were recorded using radiographs, carried out by removing an upper lung and lower lung region. These regions were rotated and interchanged and these signatures were recorded. Interactive, non statistical predictions were made on a test set of these radiographic sections with an accuracy rate of 90%. Derived features from a test set of 15 normal and 24 abnormal (2+ re-distribution) radiographs were submitted to a non-interactive stepwise discriminant analysis procedure (BMDO7M) which yielded an accuracy rate of 95% using three features. These preliminary results are encouraging and suggest that automatic processing of chest radiographs might yield estimates of the pulmonary venous pressure.