Advanced deconvolution techniques and medical radiography
Medical radiography is a process by which the internal structures of the human body are imaged using a source of x-rays. The images formed are essentially shadowgrams whose size and intensity is dependent on the geometry of the imaging system and the degree to which the structures attenuate x-ray radiation. The images are blurred because the x-ray source has a finite size, and noisy because the x-ray exposure must be kept as low as possible for the safety of the patient but which also limits the number of photons available for image formation. In such noisy environments traditional methods of Fourier deconvolution have limited appeal. In this research we apply maximum entropy methods (MEM) to some radiological images. We justify the choice of MEM over other deconvolution schemes by processing a selection of artificial images in which the blur and noise mimic the real situation but whose levels are known a priori. A hybrid MEM scheme is developed to address the shortcomings of so-called historic MEM in these situations. We initially consider images from situations in which the model point- spread function is assumed to be three-dimensionally spatially invariant, and which approximates the real situation reasonably well. One technique lends itself well to this investigation: magnification mammography. MEM is offered as a way of breaking some of the conflicting performance requirements of this technique and we explore several new system possibilities with a working MEM system in place. A more complicated blurring function is encountered in linear tomography, which uses opposing movements of the image receptor and x-ray source to generate planar images through an object. Features outside a particular focal plane are smeared to such an extent that detail within the focal plane can be very difficult to detect. With appropriate modification of our MEM technique, processed images show a significant reduction to the blurring outside the focal plane.