The analysis of medical images has been woven into the fabric of the pattern analysis and machine intelligence (PAMI) community since the earliest days of these Transactions. Initially, the efforts in this area were seen as applying pattern analysis and computer vision techniques to another interesting dataset. However, over the last two to three decades, the unique nature of the problems presented within this area of study have led to the development of a new discipline in its own right. Examples of these include: the types of image information that are acquired, the fully three-dimensional image data, the nonrigid nature of object motion and deformation, and the statistical variation of both the underlying normal and abnormal ground truth. In this paper, we look at progress in the field over the last 20 years and suggest some of the challenges that remain for the years to come.

https://hal.inria.fr/inria-00615100/document

Medical image analysis: progress over two decades and the challenges ahead

Adaptive histogram equalization (ahe) is a contrast enhancement method designed to be broadly applicable and having demonstrated effectiveness. However, slow speed and the overenhancement of noise it produces in relatively homogeneous regions are two problems. We report algorithms designed to overcome these and other concerns. These algorithms include interpolated ahe, to speed up the method on general purpose computers; a version of interpolated ahe designed to run in a few seconds on feedback processors; a version of full ahe designed to run in under one second on custom VLSI hardware; weighted ahe, designed to improve the quality of the result by emphasizing pixels' contribution to the histogram in relation to their nearness to the result pixel; and clipped ahe, designed to overcome the problem of overenhancement of noise contrast. We conclude that clipped ahe should become a method of choice in medical imaging and probably also in other areas of digital imaging, and that clipped ahe can be made adequately fast to be routinely applied in the normal display sequence.

Adaptive histogram equalization and its variations

This chapter reviews and discusses various aspects of texture analysis. The concentration is o the various methods of extracting textural features from images. The geometric, random field, fractal, and signal processing models of texture are presented. The major classes of texture processing pro lems such as segmentation, classification, and shape from texture are discussed. The possible applic tion areas of texture such as automated inspection, document processing, and remote sensing a summarized. A bibliography is provided at the end for further reading.

Texture analysis

Visual attention: The past 25 years

the of the magnitude, i.e., intensity, of the field.) It is shown that Rayleigh statistics govern the fist-order behavior of the magnitude; and the autocorrelation of the resulting image speckle is obtained by the methodof Middleton. The corresponding power spectrum follows immediately by Fourier transformation. Theoretical and experimentally determined autocorrelation functions and power spectra derived from B-scans of a scattering phantom containing many scatterers per resolution cell are presented. These functions lead naturally to the definition of the average speckle spot or cell sue, and this inturn is comparable to the resolution cell. Each independent speckle servesas a degreeof freedom that determines the number of samples of tissue available over a target.As the speckle cell size decreases this number increases in a manner predictable from the physical parameters of the cell size. However, it is found that the speckle cellis broadened, the degrees of freedom diminished, when the object structureis correlated. This yields the possibilityof deducing information about the object structure from the second-order statistics of the speckle texture, in addition to that obtainable from the fiistorder statistics.

Statistics of Speckle in Ultrasound B-Scans

A method for estimating structural properties of random media is described. The size, number density, and scattering strength of particles are estimated from an analysis of the radio frequency (rf) echo signal power spectrum. Simple correlation functions and the accurate scattering theory of Faran [J.J. Faran, J. Acoust. Soc. Am. 23, 405-418 (1951)], which includes the effects of shear waves, were used separately to model backscatter from spherical particles and thereby describe the structures of the medium. These methods were tested using both glass sphere-in-agar and polystyrene sphere-in-agar scattering media. With the appropriate correlation function, it was possible to measure glass sphere diameters with an accuracy of 20%. It was not possible to accurately estimate the size of polystyrene spheres with the simple spherical and Gaussian correlation models examined because of a significant shear wave contribution. Using the Faran scattering theory for spheres, however, the accuracy for estimating diameters was improved to 10% for both glass and polystyrene scattering media. It was possible to estimate the product of the average scattering particle number density and the average scattering strength per particle, but with lower accuracy than the size estimates. The dependence of the measurement accuracy on the inclusion of shear waves, the wavelength of sound, and medium attenuation are considered, and the implications for describing the structure of biological soft tissues are discussed.

/pdf/describing-small-scale-structure-in-random-media-using-pulse-707xxdgwi8.pdf

Describing small-scale structure in random media using pulse-echo ultrasound.

We have measured the overall statistical efficiency of human subjects discriminating the amplitude of visual pattern signals added to noisy backgrounds. By changing the noise amplitude, the amount of intrinsic noise can be estimated and allowed for. For a target containing a few cycles of a spatial sinusoid of about 5 cycles per degree, the overall statistical efficiency is as high as 0.7 +/- 0.07, and after correction for intrinsic noise, efficiency reaches 0.83 +/- 0.15. Such a high figure leaves little room for residual inefficiencies in the neural mechanisms that handle these patterns.

https://science.sciencemag.org/content/sci/214/4516/93.full.pdf

Efficiency of human visual signal discrimination.

Refinements to previous analyses of the natural correlation lengths within simple images and between images to be compounded are presented. Comparison of theoretical and experimental results show very good agreement for the case of Rayleigh scattering media: the correlation length within a simple image is comparable to the resolution cell size; the correlation length between images to be spatially compounded is comparable to, but smaller than, the transducer on array aperture; and the correlation length between images to be frequency-compounded becomes a frequency comparable to their bandwidth. Complications arising from the presence of specular scattering or due to the presence of just a few scatterers are considered. It is shown that straightforward solutions exist for either of these problems taken by itself. When they occur in combination, calibration techniques may lead to unambiguous identification of the contributions to the scattering from diffuse or incoherent scattering and from specular or coherent scattering, and to estimation of the density of diffuse scatterers. >

Fundamental correlation lengths of coherent speckle in medical ultrasonic images

Tissue signatures are obtained from the first- and second-order statistics of ultrasonic B-scan texture. Laboratory measurements and early clinical results show that the image may be segmented to discriminate between different normal tissues and to detect abnormal conditions based on a multidimensional feature space. These features describe the intrinsic backscatter properties of the tissues imaged and may be used as the basis of an automatic tissue characterization algorithm.

Analysis of ultrasound image texture via generalized rician statistics

The imaging process has two fundamental stages: detection and display. The detection stage can be quantified rigourously using Shannon's information theory. This requires the contrast scale (CS), modulation transfer function (MTF), and noise power spectrum [N(f)] to be combined into a signal-to-noise ratio (SNR). This results in two fundamental summary figures of merit: the density of noise equivalent quanta (NEQ) in the image and the information bandwidth integral (IBWI). These algorithm-independent measures are used to quantify the recording stage. The display stage is less well understood since it couples to an external observer. Several types of decision makers are treated. Examples are drawn from first and second generation CT, demonstrating that thye are nearly quantum limited for large signals, indicating how their algorithms are matched or mismatched to the geometry, and calculating the contrast-detail diagrams for those decision makers.

Robert F. Wagner

Papers

Describing small-scale structure in random media using pulse-echo ultrasound.

Efficiency of human visual signal discrimination.

Fundamental correlation lengths of coherent speckle in medical ultrasonic images

Analysis of ultrasound image texture via generalized rician statistics

Application of information theory to the assessment of computed tomography