IMAGE STAINING AND DIFFERENTIAL-DIAGNOSIS OF ULTRASOUND SCANS BASED ON THE MAHALANOBIS DISTANCE

We have previously reported on the ability to detect and discriminate among several diffuse disease states in human liver using a four-dimen sional feature space derived from the statistical physics of ultrasoun d B-scan speckle. No image resulted from this method of supervised cla ssification. In the present work the covariance matrices associated wi th each state of health or disease from that study are used as the bas is of an image staining display technique for aid in quantitative diff erential diagnosis. A state of health or disease is chosen by the clin ician: this selects the covariance matrix from the data base. A region of interest (ROI) is then scrolled through an abdominal B-scan. For e ach position of the ROI a point in the four-dimensional feature space is calculated. A natural measure of the distance of this point from th e center of mass (multivariate mean) of the disease class is calculate d in terms of the covariance matrix of this class; this measure is the Mahalanobis distance. The confidence level for acceptance or rejectio n of the hypothesized disease class is obtained from the probability d istribution of this distance, the T2 probability law. This confidence level is color coded and used as a color stain that overlays the origi nal scan at that position. The variability of the calculated features is studied as a function of ROI size, or the spatial resolution of the color coded image, and it is found that for an ROI in the neighborhoo d of 4 cm2 most of the variability due to the finite number of indepen dent samples (speckles) is averaged out, leaving the ''noise floor'' a ssociated with inter- and intra-patient variability. ROIs on the order of 1 cm2 may result with technical advances in B-scan resolution. A s mall number of points on organ boundaries are entered by the user, to fit with arcs of ellipses to be used to switch between organ (liver an d kidney) data bases as the ROI encounters the boundary. By selecting in turn various state-of-health or state-of-disease databases, such im ages of confidence levels may be used for quantitative differential di agnosis. The method is not limited to ultrasound, being applicable in principle to features obtained from any modality or multimodality comb ination.