Predictive Deconvolution and Hybrid Feature Selection for Computer-Aided Detection of Prostate Cancer

Abstract

Computer-aided detection (CAD) schemes are decision making support tools, useful to overcome limitations of problematic clinical procedures. Trans-rectal ultrasound image based CAD would be extremely important to support prostate cancer diagnosis. An effective approach to realize a CAD scheme for this purpose is described in this work, employing a multi-feature kernel classification model based on generalized discriminant analysis. The mutual information of feature value and tissue pathological state is used to select features essential for tissue characterization. System-dependent effects are reduced through predictive deconvolution of the acquired radio-frequency signals. A clinical study, performed on ground truth images from biopsy findings, provides a comparison of the classification model applied before and after deconvolution, showing in the latter case a significant gain in accuracy and area under the receiver operating characteristic curve.

Soft-label reinforced rtCAB for guided prostate tissue sampling

“…This system has been designed to guide tissue sampling to areas with the highest probability of PCa and to provide real-time assistance to physician during biopsy by implementation with the CUDA parallel processing platform. Since the effectiveness of using different types of features extracted from the TRUS data has been shown in several studies [3][4], in this research features of different natures are employed in the structure of an ensemble classification model and partial labels assigned to some of the data are reconsidered by a label assignment phase. The label assignment method provides soft or crisp class labels for patterns with uncertain labels and in this way the size of the training set can be increased.…”

Continuous-domain ARMA modeling for ultrasound tissue characterization

“…In particular, the statistical modeling of discrete ultrasound signals has been widely employed to derive interesting properties from estimated model parameters that are useful in the context of tissue characterization. Among them, shift-variant autoregressive moving-average (ARMA) processes were found suitable to model the ultrasound signals and images and to make the acquisition more robust in its estimation of the uncorrupted tissue response [1,2]. More recently, 2D ARMA modeling was proposed to improve computer-aided detection of breast tumors [3,4].…”

A novel NMF-based DWI CAD framework for prostate cancer.

“…From the pixel/voxel intensities, several features can be obtained. For example, Gaussian statistics (mean and standard deviation) of pixel/voxels intensities are used as features in several TRUS studies [151][152][153][154][155]. The Nakagami distribution has also been used for extracting features from pixel intensities in various studies [153,154].…”

“…Examples of these include wavelet coefficients [160] and their polynomial fitting [160], which were used in [153][154][155], the autocorrelation coefficients [151], and a tumor's shape metric [151]. In addition, fractal texture features [161] and spectral features [162] were also used in prostate CAD systems, such as in [155] and [153][154][155], respectively. Another feature utilized is the total least square estimation of signal parameters [157], which is estimated via rotational invariance techniques (TLS-ESPRIT) [163,164].…”

“…For example, Maggio et al [153,154] proposed an automated CAD system that utilized the tissue intensities, features extracted using the Nakagami distribution, Haralick textual features, and Unser textual features to classify prostate tissues as benign or malignant. Using an LDA classifier, this system had a sensitivity of 75±9%, a specificity of 93±2%, an accuracy of 93±2%, and an AUC of 0.95±0.02.…”