Image restoration in radioisotope imaging systems

Optimization by simulated annealing of three-dimensional conformal treatment planning for radiation fields defined by a multileaf collimator

Three-dimensional conformal radiotherapy may be achieved by using a combination of geometrically shaped radiation fields from different orientations around the patient. A convenient method to shape the fields is to use a multileaf collimator. These fields are shaped to the beam's-eye-view of the target volume, at each orientation of the collimator, and may also encompass sensitive structure, i.e. organs at risk, if the target region has concavities in its outline within which such structure may reside. The term 'conformal therapy' is used in this paper to mean tailoring the high dose volume to the target volume whilst minimising dose to other normal structures (organs at risk) which may be irradiated by the treatment fields, shaped by a multileaf collimator. The question then arises of the optimum distribution of beam weights to apply to the fields to minimise dose to organs at risk whilst aiming towards a uniform dose distribution in the target volume. This paper provides a method of optimizing the choice of beamweights to achieve this. The method is based on the well known optimization technique of simulated annealing. Either an optimal set of beamweights, one weight per field, is generated or the intensity may be spatially modulated across the field at each orientation (two weights per field) depending on whether there is just target volume or both target volume and volume containing organs at risk in the line of sight. It is shown that the dose matrix resulting from the latter optimization is closer to the dose prescription than that obtained by using either an optimal set of single weights per field or uniform beamweights.

SPECT volume measurement using an automatic threshold selection method combined with a V filter

The volume measurement of various organs by single photon emission computed tomography (SPECT) has generally been performed by summing up the voxels lying within the outer edge of the organ determined by thresholding as a function of the maximum reconstructed counts within a volume of interest. The optimum threshold level, however, is influenced by various factors. To eliminate this problem, Mortelmans et al. (1986) proposed to use an automatic threshold selection method (ATSM) based on discriminant criteria and to correct the errors induced by the imperfect system response using the method based on linear regression analysis, which has some problems in practical application. We tried to use ATSM combined with a V filter. The correlation coefficient (r) and the regression equation between the true (x) and the calculated volume (y) obtained by experimentation using 22 phantoms with a volume ranging between 26 ml and 380 ml were as follows: r = 0.994 and y(ml) = 1.015 x +1.09 using ATSM combined with a V filter, while r = 0.993 and y(ml) = 1.110 x +30.55 using only ATSM. The relative error between the true and the calculated volumes decreased significantly (P less than 0.01) from 37.7% +/- 26.9% to 5.5% +/- 4.2%. The results suggest that ATSM combined with a V filter is useful for correcting the errors caused by an imperfect system response, and is available and reliable for SPECT volume determination even when the organ volume is small.

Multiple photon coincidence tomography

Coincidence scanning devices that measure the distribution of radioisotopes emitting multiple photons in nuclear cascade decays offer a possible supplementary approach to tomography in nuclear medicine. Design factors that serve to determine resolution, sensitivity, and statistical noise for the multiphoton coincidence linear scanner (MCLS), the total organ kinetic imaging monitor (TOKIM), and related systems have been well studied. Focused collimator coincidence scanner (FCCS) systems are capable of extremely high resolution--spherical cold lesions of less than 0.2 cu cm volume being easily detectable. Although FCCS scan speeds are too slow for imaging of large organs, scan times for small organs or for the rescanning of suspicious or ambigous regions appearing on conventional scans are well within practical clinical limits. In view of recent developments in the on-site cyclotron production of short-lived radioisotopes and the current interest in high resolution localization of neurologic receptors in vivo, FCCS systems may prove also to be of value in basic neurophysiologic studies.

Constrained optimisation and its application to scintigraphy

Any instrument that records data will produce a distorted or degraded version of the input signal. For example, imaging equipment will produce a blurred image of the object, and in the case of scintigraphic imaging the blurs may be comparable to the size of the physiological structures being investigated. The process of refocusing is called, in mathematical terms, deconvolution. In scintigraphy most workers have implemented deconvolution based on Wiener filtering. The shortcomings of the Wiener method when applied to scintigraphy are identified and an alternative approach to deconvolution using constrained optimisation is presented. The mathematical derivation of the new restoring filter is given in terms of the discrete Fourier transform. The practical implementation of the method on a digital computer is described and, as an example, the result of deconvolving a brain scan is presented.

The analysis of radioisotope brain scans by comparison with normal patterns of uptake

A previously described method for the computer analysis of radioisotope brain scans has been further developed. A patient scan is matched with a series of "normal" scans to find the closest fit and the "normal" scan is then subtracted from the patient scan. The results are displayed in terms of the standard deviation from the normal. In a clinical trial the computer assessments were found to be less accurate than those of two groups of observers. One reason for these disappointing results is related to the simple criteria used to choose and fit the best "normal" to a particular patient scan. A second reason is the optimization of the scanner/photoscan combination to 99Tcm pertechnetate as a brain scanning agent which results in patient-to-patient variation becoming more important. This reduces the potential of a simple numerical analysis to improve results.