Digital Processing Of Medical Images For Computer-Aided Diagnosis
Maryellen Lissak Giger, Kunio Doi, Shigehiko Katsuragawa, Kenneth R. Hoffmann, Heber MacMahon · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1989
Maryellen L. Giger, Kunio Doi, Shigehiko Katsuragawa, Kenneth R. Hoffmann, Heber MacMahonKurt Rossmann Laboratories for Radiologic Image ResearchDepartment of Radiology, The University of ChicagoChicago, Illinois 60637ABSTRACTVarious image processing techniques and computer -aided diagnostic schemes for digitalradiographic images are reviewed. Computer -aided schemes, which are in development at the KurtRossmann Laboratories for Radiologic Image Research of the University of Chicago, arc in the fields ofchest radiography, angiography, and mammography.INTRODUCTIONThe interpretation of medical images by radiologists is a rather subjective process during whichvisible abnormalities may be missed. These missed diagnoses may be due to the camouflaging effect of thesurrounding anatomic background, the subjective and varying decision criteria used by radiologists, orvarious distractions present in the clinical situation (1 -5). If computerized methods can be developed todetect and /or classify abnormalities in digital images, then it is expected that radiologists, alerted to possibleabnormal locations by the computer, would incur fewer false -negative diagnoses. Radiologists would acceptor reject the computer -reported suggestions, before making the final diagnostic decisions.This paper reviews some of the computer -aided diagnosis (CAD) schemes under development andinvestigation in the Kurt Rossmann Laboratories for Radiologie Image Research at the University ofChicago. The CAD schemes discussed here are in the fields of chest radiography and angiography.Computerized schemes in radiography can be divided broadly into two parts: (1) the detection andlocalization of a possible abnormality (computer vision) and (2) the merging of image features into adiagnostic decision regarding the presence or absence of the abnormality (expert system.) The developmentof computer -aided diagnosis schemes requires a priori information about the medical image, and knowledgeof computer processing and feature -extraction techniques. The required a priori knowledge about the specificmedical image to be analyzed includes the physical imaging properties of the acquisition system, andmorphological information of the abnormality in question along with its associated anatomic background(i.e., a database). Figure 1 illustrates a general scheme for CAD. Examples are given for the two chestradiography schemes to be discussed in this paper. The components of a computerized scheme tend tooverlap, however general descriptions (6 -8) of each stage will be attempted here. Point and spatialprocessing techniques can be employed to reduce noise and other interfering patterns or to enhance thefeatures to he later extracted. Segmentation of an image involves the separation of the image into regionsof similar attributes. Basically, a segmentor only subdivides the image and does not try to recognize theindividual segments. Feature extraction entails the recognition of specific characteristics of the individualsegments. Finally in the classification stage, feature descriptors can be analyzed and interpreted in order toclassify the input pattern into one of several categories, e.g., normal or abnormal.CHEST RADIOGRAPHY