An Adaptive Learning Model for Multiscale Texture Features in Polyp Classification via Computed Tomographic Colonography.

Objective: As an effective lesion heterogeneity depiction, texture information extracted from computed tomography has become increasingly important in polyp classification. However, variation and redundancy among multiple texture descriptors render a challenging task of integrating them into a general characterization. Considering these two problems, this work proposes an adaptive learning model to integrate multi-scale texture features. Methods: To mitigate feature variation, the whole feature set is geometrically split into several independent subsets that are ranked by a learning evaluation measure after preliminary classifications. To reduce feature redundancy, a bottom-up hierarchical learning framework is proposed to ensure monotonic increase of classification performance while integrating these ranked sets selectively. Two types of classifiers, traditional (random forest + support vector machine)- and convolutional neural network (CNN)-based, are employed to perform the polyp classification under the proposed framework with extended Haralick measures and gray-level co-occurrence matrix (GLCM) as inputs, respectively. Experimental results are based on a retrospective dataset of 63 polyp masses (defined as greater than 3 cm in largest diameter), including 32 adenocarcinomas and 31 benign adenomas, from adult patients undergoing first-time computed tomography colonography and who had corresponding histopathology of the detected masses. Results: We evaluate the performance of the proposed models by the area under the curve (AUC) of the receiver operating characteristic curve. The proposed models show encouraging performances of an AUC score of 0.925 with the traditional classification method and an AUC score of 0.902 with CNN. The proposed adaptive learning framework significantly outperforms nine well-established classification methods, including six traditional methods and three deep learning ones with a large margin. Conclusions: The proposed adaptive learning model can combat the challenges of feature variation through a multiscale grouping of feature inputs, and the feature redundancy through a hierarchal sorting of these feature groups. The improved classification performance against comparative models demonstrated the feasibility and utility of this adaptive learning procedure for feature integration.

Examining feature extraction and classification modules in machine learning for diagnosis of low-dose computed tomographic screening-detected in vivo lesions.

Purpose: Medical imaging-based machine learning (ML) for computer-aided diagnosis of in vivo lesions consists of two basic components or modules of (i) feature extraction from non-invasively acquired medical images and (ii) feature classification for prediction of malignancy of lesions detected or localized in the medical images. This study investigates their individual performances for diagnosis of low-dose computed tomography (CT) screening-detected lesions of pulmonary nodules and colorectal polyps. Approach: Three feature extraction methods were investigated. One uses the mathematical descriptor of gray-level co-occurrence image texture measure to extract the Haralick image texture features (HFs). One uses the convolutional neural network (CNN) architecture to extract deep learning (DL) image abstractive features (DFs). The third one uses the interactions between lesion tissues and X-ray energy of CT to extract tissue-energy specific characteristic features (TFs). All the above three categories of extracted features were classified by the random forest (RF) classifier with comparison to the DL-CNN method, which reads the images, extracts the DFs, and classifies the DFs in an end-to-end manner. The ML diagnosis of lesions or prediction of lesion malignancy was measured by the area under the receiver operating characteristic curve (AUC). Three lesion image datasets were used. The lesions' tissue pathological reports were used as the learning labels. Results: Experiments on the three datasets produced AUC values of 0.724 to 0.878 for the HFs, 0.652 to 0.965 for the DFs, and 0.985 to 0.996 for the TFs, compared to the DL-CNN of 0.694 to 0.964. These experimental outcomes indicate that the RF classifier performed comparably to the DL-CNN classification module and the extraction of tissue-energy specific characteristic features dramatically improved AUC value. Conclusions: The feature extraction module is more important than the feature classification module. Extraction of tissue-energy specific characteristic features is more important than extraction of image abstractive and characteristic features.

Lesion Classification by Model-Based Feature Extraction: A Differential Affine Invariant Model of Soft Tissue Elasticity in CT Images.

The elasticity of soft tissues has been widely considered a characteristic property for differentiation of healthy and lesions and, therefore, motivated the development of several elasticity imaging modalities, for example, ultrasound elastography, magnetic resonance elastography, and optical coherence elastography to directly measure the tissue elasticity. This paper proposes an alternative approach of modeling the elasticity for prior knowledge-based extraction of tissue elastic characteristic features for machine learning (ML) lesion classification using computed tomography (CT) imaging modality. The model describes a dynamic non-rigid (or elastic) soft tissue deformation in differential manifold to mimic the tissues' elasticity under wave fluctuation in vivo. Based on the model, a local deformation invariant is formulated using the 1st and 2nd order derivatives of the lesion volumetric CT image and used to generate elastic feature map of the lesion volume. From the feature map, tissue elastic features are extracted and fed to ML to perform lesion classification. Two pathologically proven image datasets of colon polyps and lung nodules were used to test the modeling strategy. The outcomes reached the score of area under the curve of receiver operating characteristics of 94.2% for the polyps and 87.4% for the nodules, resulting in an average gain of 5 to 20% over several existing state-of-the-art image feature-based lesion classification methods. The gain demonstrates the importance of extracting tissue characteristic features for lesion classification, instead of extracting image features, which can include various image artifacts and may vary for different protocols in image acquisition and different imaging modalities.

Exploring Dual-Energy CT Spectral Information for Machine Learning-Driven Lesion Diagnosis in Pre-Log Domain.

In this study, we proposed a computer-aided diagnosis (CADx) framework under dual-energy spectral CT (DECT), which operates directly on the transmission data in the pre-log domain, called CADxDE, to explore the spectral information for lesion diagnosis. The CADxDE includes material identification and machine learning (ML) based CADx. Benefits from DECT's capability of performing virtual monoenergetic imaging with the identified materials, the responses of different tissue types (e.g., muscle, water, and fat) in lesions at each energy can be explored by ML for CADx. Without losing essential factors in the DECT scan, a pre-log domain model-based iterative reconstruction is adopted to obtain decomposed material images, which are then used to generate the virtual monoenergetic images (VMIs) at selected n energies. While these VMIs have the same anatomy, their contrast distribution patterns contain rich information along with the n energies for tissue characterization. Thus, a corresponding ML-based CADx is developed to exploit the energy-enhanced tissue features for differentiating malignant from benign lesions. Specifically, an original image-driven multi-channel three-dimensional convolutional neural network (CNN) and extracted lesion feature-based ML CADx methods are developed to show the feasibility of CADxDE. Results from three pathologically proven clinical datasets showed 4.01% to 14.25% higher AUC (area under the receiver operating characteristic curve) scores than the scores of both the conventional DECT data (high and low energy spectrum separately) and the conventional CT data. The mean gain >9.13% in AUC scores indicated that the energy spectral-enhanced tissue features from CADxDE have great potential to improve lesion diagnosis performance.

Vector textures derived from higher order derivative domains for classification of colorectal polyps.

Textures have become widely adopted as an essential tool for lesion detection and classification through analysis of the lesion heterogeneities. In this study, higher order derivative images are being employed to combat the challenge of the poor contrast across similar tissue types among certain imaging modalities. To make good use of the derivative information, a novel concept of vector texture is firstly introduced to construct and extract several types of polyp descriptors. Two widely used differential operators, i.e., the gradient operator and Hessian operator, are utilized to generate the first and second order derivative images. These derivative volumetric images are used to produce two angle-based and two vector-based (including both angle and magnitude) textures. Next, a vector-based co-occurrence matrix is proposed to extract texture features which are fed to a random forest classifier to perform polyp classifications. To evaluate the performance of our method, experiments are implemented over a private colorectal polyp dataset obtained from computed tomographic colonography. We compare our method with four existing state-of-the-art methods and find that our method can outperform those competing methods over 4%-13% evaluated by the area under the receiver operating characteristics curves.

A dynamic lesion model for differentiation of malignant and benign pathologies.

Malignant lesions have a high tendency to invade their surrounding environment compared to benign ones. This paper proposes a dynamic lesion model and explores the 2nd order derivatives at each image voxel, which reflect the rate of change of image intensity, as a quantitative measure of the tendency. The 2nd order derivatives at each image voxel are usually represented by the Hessian matrix, but it is difficult to quantify a matrix field (or image) through the lesion space as a measure of the tendency. We conjecture that the three eigenvalues contain important information of the Hessian matrix and are chosen as the surrogate representation of the Hessian matrix. By treating the three eigenvalues as a vector, called Hessian vector, which is defined in a local coordinate formed by three orthogonal Hessian eigenvectors and further adapting the gray level occurrence computing method to extract the vector texture descriptors (or measures) from the Hessian vector, a quantitative presentation for the dynamic lesion model is completed. The vector texture descriptors were applied to differentiate malignant from benign lesions from two pathologically proven datasets: colon polyps and lung nodules. The classification results not only outperform four state-of-the-art methods but also three radiologist experts.

3D-GLCM CNN: A 3-Dimensional Gray-Level Co-Occurrence Matrix-Based CNN Model for Polyp Classification via CT Colonography.

Accurately classifying colorectal polyps, or differentiating malignant from benign ones, has a significant clinical impact on early detection and identifying optimal treatment of colorectal cancer. Convolution neural network (CNN) has shown great potential in recognizing different objects (e.g. human faces) from multiple slice (or color) images, a task similar to the polyp differentiation, given a large learning database. This study explores the potential of CNN learning from multiple slice (or feature) images to differentiate malignant from benign polyps from a relatively small database with pathological ground truth, including 32 malignant and 31 benign polyps represented by volumetric computed tomographic (CT) images. The feature image in this investigation is the gray-level co-occurrence matrix (GLCM). For each volumetric polyp, there are 13 GLCMs, computed from each of the 13 directions through the polyp volume. For comparison purpose, the CNN learning is also applied to the multi-slice CT images of the volumetric polyps. The comparison study is further extended to include Random Forest (RF) classification of the Haralick texture features (derived from the GLCMs). From the relatively small database, this study achieved scores of 0.91/0.93 (two-fold/leave-one-out evaluations) AUC (area under curve of the receiver operating characteristics) by using the CNN on the GLCMs, while the RF reached 0.84/0.86 AUC on the Haralick features and the CNN rendered 0.79/0.80 AUC on the multiple-slice CT images. The presented CNN learning from the GLCMs can relieve the challenge associated with relatively small database, improve the classification performance over the CNN on the raw CT images and the RF on the Haralick features, and have the potential to perform the clinical task of differentiating malignant from benign polyps with pathological ground truth.

Multi-scale characterizations of colon polyps via computed tomographic colonography.

Texture features have played an essential role in the field of medical imaging for computer-aided diagnosis. The gray-level co-occurrence matrix (GLCM)-based texture descriptor has emerged to become one of the most successful feature sets for these applications. This study aims to increase the potential of these features by introducing multi-scale analysis into the construction of GLCM texture descriptor. In this study, we first introduce a new parameter - stride, to explore the definition of GLCM. Then we propose three multi-scaling GLCM models according to its three parameters, (1) learning model by multiple displacements, (2) learning model by multiple strides (LMS), and (3) learning model by multiple angles. These models increase the texture information by introducing more texture patterns and mitigate direction sparsity and dense sampling problems presented in the traditional Haralick model. To further analyze the three parameters, we test the three models by performing classification on a dataset of 63 large polyp masses obtained from computed tomography colonoscopy consisting of 32 adenocarcinomas and 31 benign adenomas. Finally, the proposed methods are compared to several typical GLCM-texture descriptors and one deep learning model. LMS obtains the highest performance and enhances the prediction power to 0.9450 with standard deviation 0.0285 by area under the curve of receiver operating characteristics score which is a significant improvement.

Multilayer feature selection method for polyp classification via computed tomographic colonography.

Polyp classification is a feature selection and clustering process. Picking the most effective features from multiple polyp descriptors without redundant information is a great challenge in this procedure. We propose a multilayer feature selection method to construct an optimized descriptor for polyp classification with a feature-grouping strategy in a hierarchical framework. First, the proposed method makes good use of image metrics, such as intensity, gradient, and curvature, to divide their corresponding polyp descriptors into several feature groups, which are the preliminary units of this method. Then each preliminary unit generates two ranked descriptors, i.e., their optimized variable groups (OVGs) and preliminary classification measurements. Next, a feature dividing-merging (FDM) algorithm is designed to perform feature merging operation hierarchically and iteratively. Unlike traditional feature selection methods, the proposed FDM algorithm includes two steps for feature dividing and feature merging. At each layer, feature dividing selects the OVG with the highest area under the receiver operating characteristic curve (AUC) as the baseline while other descriptors are treated as its complements. In the fusion step, the FDM merges some variables with gains into the baseline from the complementary descriptors iteratively on every layer until the final descriptor is obtained. This proposed model (including the forward step algorithm and the FDM algorithm) is a greedy method that guarantees clustering monotonicity of all OVGs from the bottom to the top layer. In our experiments, all the selected results from each layer are reported by both graphical illustration and data analysis. Performance of the proposed method is compared to five existing classification methods by a polyp database of 63 samples with pathological reports. The experimental results show that our proposed method outperforms other methods by 4% to 23% gains in terms of AUC scores.

A Feasibility Study of Extracting Tissue Textures From a Previous Full-Dose CT Database as Prior Knowledge for Bayesian Reconstruction of Current Low-Dose CT Images.

Markov random field (MRF) has been widely used to incorporate a priori knowledge as penalty or regularizer to preserve edge sharpness while smoothing the region enclosed by the edge for pieces-wise smooth image reconstruction. In our earlier study, we proposed a type of MRF reconstruction method for low-dose CT (LdCT) scans using tissue-specific textures extracted from the same patient's previous full-dose CT (FdCT) scans as prior knowledge. It showed advantages in clinical applications. This paper aims to remove the constraint of using previous data of the same patient. We investigated the feasibility of extracting the tissue-specific MRF textures from an FdCT database to reconstruct a LdCT image of another patient. This feasibility study was carried out by experiments designed as follows. We constructed a tissue-specific MRF-texture database from 3990 FdCT scan slices of 133 patients who were scheduled for lung nodule biopsy. Each patient had one FdCT scan (120 kVp/100 mAs) and one LdCT scan (120 kVp/20 mAs) prior to biopsy procedure. When reconstructing the LdCT image of one patient among the 133 patients, we ranked the closeness of the MRF-textures from the other 132 patients saved in the database and used them as the a prior knowledge. Then, we evaluated the reconstructed image quality using Haralick texture measures. For any patient within our database, we found more than eighteen patients' FdCT MRF texures can be used without noticeably changing the Haralick texture measures on the lung nodules (to be biopsied). These experimental outcomes indicate it is promising that a sizable FdCT texture database could be used to enhance Bayesian reconstructions of any incoming LdCT scans.