CRITICAL AND EMERGING ISSUES IN DIAGNOSTIC IMAGING: DIMENSIONALITY OF THE FEATURE SPACE.

X-ray, nuclear magnetic resonance, and ultrasonic imaging in medicine are capable of contributing, respectively, two, three or four, and four or five physical measurements of features for tissue characterization. These physical quantities are reviewed with regard to the stochastic and deterministic errors in their measuremnts. The nature of the problem of dimensionality reduction is illustrated in the combined feature space by drawing on some examples from diagnostic ultrasound. A summary of the methods for assessing the limits to sensitivity of imaging systems and their quantification is also given. This includes finding the relative (and absolute) contributions to the image fluctuations, or noise, due to natural fluctuations in the signals themselves, systematic effects or variable calibration of the detection system, and the natural spread in signal strengths due to the variability in the patient population.