Volumetric Textural Analysis of Colorectal Masses at CT Colonography: Differentiating Benign versus Malignant Pathology and Comparison with Human Reader Performance

Radiomics-based machine learning in the differentiation of benign and malignant bowel wall thickening radiomics in bowel wall thickening

PURPOSE

To distinguish malignant and benign bowel wall thickening (BWT) by using computed tomography (CT) texture features based on machine learning (ML) models and to compare its success with the clinical model and combined model.

METHODS

One hundred twenty-two patients with BWT identified on contrast-enhanced abdominal CT and underwent colonoscopy were included in this retrospective study. Texture features were extracted from CT images using LifeX software. Feature selection and reduction were performed using the Least Absolute Shrinkage and Selection Operator (LASSO). Six radiomic features were selected with LASSO. In the clinical model, six features (age, gender, thickness, fat stranding, symmetry, and lymph node) were included. Six radiomic and six clinical features were used in the combined model. Classification was done using two machine learning algorithms: Support Vector Machine (SVM) and Logistic Regression (LR). The data sets were divided into 80% training set and 20% test set. Then, training took place with all three datasets. The model's success was tested with the test set consisting of features not used during training.

RESULTS

In the training set, the combined model had the best performance with the area under the curve (AUC) value of 0.99 for SVM and 0.95 for LR. In the radiomic-derived model, the AUC value is 0.87 in SVM and 0.79 in LR. In the clinical model, SVM made this distinction with 0.95 AUC and LR with 0.92 AUC value. In the test set, the classifier with the highest success distinguishing malignant wall thickening is SVM in the radiomic-derived model with an AUC value of 0.90. In other models, the AUC value is in the range of 0.75-0.86, and the accuracy values are in the range of 0.72-0.84.

CONCLUSION

In conclusion, radiomic-based machine learning has shown high success in distinguishing malignant and benign BWT and may improve diagnostic accuracy compared to clinical features only. The results of our study may help ensure early diagnosis and treatment of colorectal cancers by facilitating the recognition of malignant BWT.

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.

Appendiceal Inversion Presenting as a Cecal Polypoid Mass on Screening Colonoscopy: A Case Report and Review of Available Diagnostic Adjuncts to Differentiate Benign From Malignant Colorectal Pathology

Appendiceal inversion is uncommon. It may be a benign finding or seen in association with malignant pathology. When detected, it masquerades as a cecal polyp which poses a diagnostic dilemma with malignancy in the differential. In this report, we highlight a case of a 51-year-old patient with an extensive surgical history as a newborn in the setting of omphalocele and intestinal malrotation, who was found to have a 4 cm cecal polypoid growth on screening colonoscopy. He underwent a cecectomy for tissue diagnosis. Ultimately, the polyp was found to be an inverted appendix without evidence of malignancy. Currently, suspicious colorectal lesions which cannot be removed by polypectomy are primarily addressed with surgical excision. We reviewed the literature for available diagnostic adjuncts to better differentiate benign from malignant colorectal pathology. The application of advanced imaging and molecular technology will allow for improved diagnostic accuracy and subsequent operative planning.

TMSF-Net: Multi-series fusion network with treeconnect for colorectal tumor segmentation

PURPOSE

Colorectal tumors are common clinical diseases. Automatic segmentation of colorectal tumors captured in computed tomography (CT) images can provide numerous possibilities for computer-assisted treatment. Obtaining large datasets is expensive, and completing labeling is time- and manpower-consuming. To solve the challenge using a limited pathological dataset, this paper proposes a multi-series fusion network with treeconnect (TMSF-Net), which can automatically achieve colorectal tumor segmentation using CT images.

METHODS

To drive the TMSF-Net, three-series enhanced CT images were collected from all patients to improve the data characteristics. In the TMSF-Net, the coding path was designed as a three-branch structure to realize the feature extraction of the different series. Subsequently, the three branches were merged to start the feature analysis in the decoding path. To achieve the objective of feature fusion, different layers in the decoding path fused feature maps from the upper layer in the encoding path to achieve a cross-scale fusion. In addition, to reduce the problem of parameter redundancy, this study adopted a three-dimensional treeconnect to complete data connection on three branches.

RESULTS

A total of 22 cases were conducted by ablation and comparative experiments to test the TMSF-Net. The results showed that the TMSF-Net can improve the network performance by multiseries fusion, and its expressiveness is better than many classic networks.

CONCLUSION

The TMSF-Net is a many-to-one structure network, which can enhance the network learning ability and improve the analysis of potential features. Therefore, it yields good results in colorectal tumor segmentation. It can provide a new direction for neural network models based on feature fusion.

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.

Computed tomography colonography versus colonoscopy for detection of colorectal cancer: a diagnostic performance study

BACKGROUND

Colonoscopy is the reference standard for the detection of colorectal cancer but it is an invasive technique and has the risk of bowel perforation and bleeding. Unlike colonoscopy, sedation is not required in computed tomography colonography and requires additional reassurance endoscopy. The objectives of the study were to compare the diagnostic performance of computed tomography colonography against colonoscopy for a diagnosis of colorectal cancer.

METHODS

Data regarding any polyp ≥10 mm diameter (ø) and < 10 mm ø but suspicious polyps of computed tomography colonography (n = 318), colonoscopy (n = 318), and surgical pathology (n = 77) for symptomatic colorectal cancer patients were collected and analyzed. Lesion ulceration, extramural invasion, and/ or lesion shouldering was considered as a suspicious polyp. Beneficial scores for decision making of curative surgeries were evaluated for each modality. The cost of diagnosis of colorectal cancer was also evaluated.

RESULTS

Either of diagnosis showed polyps ≥10 mm ø in 27 patients and polyps of 50 patients were < 10 mm ø but suspicious. Therefore, a total of 77 patients were subjected to surgery. With respect to surgical pathology, sensitivities for computed tomographic colonography and colonoscopy were 0.961 and 0.831. For detection of ≥10 mm ø polyp, benefit score for computed tomographic colonography and colonoscopy were 0-0.906 diagnostic confidence and 0.035-0.5 diagnostic confidence. For polyps, ≥ 10 mm ø but not too many large polyps, colonoscopy had the risk of underdiagnosis. For < 10 mm ø but suspicious polyps, < 0.6 mm ø and < 2.2 mm ⌀ polyps could not be detected by computed tomographic colonography and colonoscopy, respectively. The computed tomographic colonography had less cost than colonoscopy (1345 ± 135 ¥/ patient vs. 1715 ± 241 ¥/ patient, p < 0.0001) for diagnosis of colorectal cancer.

CONCLUSION

Computed tomographic colonography would be a non-inferior alternative than colonoscopy for a diagnosis of colorectal cancer.

LEVEL OF EVIDENCE

III.

The feasibility of differentiating colorectal cancer from normal and inflammatory thickening colon wall using CT texture analysis

To investigate the diagnostic value of texture analysis (TA) for differentiating between colorectal cancer (CRC), colonic lesions caused by inflammatory bowel disease (IBD), and normal thickened colon wall (NTC) on computed tomography (CT) and assess which scanning phase has the highest differential diagnostic value. In all, 107 patients with CRC, 113 IBD patients with colonic lesions, and 96 participants with NTC were retrospectively enrolled. All subjects underwent multiphase CT examination, including pre-contrast phase (PCP), arterial phase (AP), and portal venous phase (PVP) scans. Based on these images, classification by TA and visual classification by radiologists were performed to discriminate among the three tissue types. The performance of TA and visual classification was compared. Precise TA classification results (error, 2.03–12.48%) were acquired by nonlinear discriminant analysis for CRC, IBD and NTC, regardless of phase or feature selection. PVP images showed a better ability to discriminate the three tissues by comprising the three scanning phases. TA showed significantly better performance in discriminating CRC, IBD and NTC than visual classification for residents, but there was no significant difference in classification between TA and experienced radiologists. TA could provide useful quantitative information for the differentiation of CRC, IBD and NTC on CT, particularly in PVP images.

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.