Change In Feature Values Research Articles

Icon similarity model based on cognition and deep learning

Human-computer cooperation guided by natural interaction, intelligent interaction, and human–computer integration is gradually becoming a new trend in human–computer interfaces. An icon is an indispensable pictographic symbol in an interface that can convey pivotal semantics between humans and computers. Research on similar icons’ cognition in humans and the discrimination of computers can reduce misunderstandings and facilitate transparent cooperation. Therefore, this research focuses on images of icons, extracted contours, and four features, including the curvature, proportion, orientation, and line of the contour, step by step. By manipulating the feature value change to obtain 360 similar icons, a cognitive experiment was conducted with 25 participants to explore the boundary values of the feature dimensions that cause different levels of similarity. Its boundary values were applied to deep learning to train a discrimination algorithm model that included 1500 similar icons. This dataset was used to train a Siamese neural network using a 16-layer network branch of a visual geometry group. The training process used stochastic gradient descent. This method of combining human cognition and deep learning technology is meaningful for establishing a consensus on icon semantics, including content and emotions, by outputting similarity levels and values. Taking icon similarity discrimination as an example, this study explored the analysis and simulation methods of computer vision for human visual cognition. The accuracy evaluated is 90.82%. The precision was evaluated as 90% for high, 80.65% for medium, and 97.30% for low. Recall was evaluated as 100% for high, 89.29% for medium, and 83.72% for low. It has been verified that it can compensate for fuzzy cognition in humans and enable computers to cooperate efficiently.

A Multilayer Perceptron-Based Spherical Visual Compass Using Global Features.

This paper presents a visual compass method utilizing global features, specifically spherical moments. One of the primary challenges faced by photometric methods employing global features is the variation in the image caused by the appearance and disappearance of regions within the camera's field of view as it moves. Additionally, modeling the impact of translational motion on the values of global features poses a significant challenge, as it is dependent on scene depths, particularly for non-planar scenes. To address these issues, this paper combines the utilization of image masks to mitigate abrupt changes in global feature values and the application of neural networks to tackle the modeling challenge posed by translational motion. By employing masks at various locations within the image, multiple estimations of rotation corresponding to the motion of each selected region can be obtained. Our contribution lies in offering a rapid method for implementing numerous masks on the image with real-time inference speed, rendering it suitable for embedded robot applications. Extensive experiments have been conducted on both real-world and synthetic datasets generated using Blender. The results obtained validate the accuracy, robustness, and real-time performance of the proposed method compared to a state-of-the-art method.

Marginal effects for non-linear prediction functions

Beta coefficients for linear regression models represent the ideal form of an interpretable feature effect. However, for non-linear models such as generalized linear models, the estimated coefficients cannot be interpreted as a direct feature effect on the predicted outcome. Hence, marginal effects are typically used as approximations for feature effects, either as derivatives of the prediction function or forward differences in prediction due to changes in feature values. While marginal effects are commonly used in many scientific fields, they have not yet been adopted as a general model-agnostic interpretation method for machine learning models. This may stem from the ambiguity surrounding marginal effects and their inability to deal with the non-linearities found in black box models. We introduce a unified definition of forward marginal effects (FMEs) that includes univariate and multivariate, as well as continuous, categorical, and mixed-type features. To account for the non-linearity of prediction functions, we introduce a non-linearity measure for FMEs. Furthermore, we argue against summarizing feature effects of a non-linear prediction function in a single metric such as the average marginal effect. Instead, we propose to average homogeneous FMEs within population subgroups, which serve as conditional feature effect estimates.

Brain Tumor Prediction using Adaptive Connected Component based GLCM and SVM Method

A crucial stage in the diagnosis of brain disorders using magnetic resonance images is feature extraction. The feature extraction procedure is used to reduce the amount of the picture data by removing the necessary information from the segmented image. The segmentation strategy and features that are extracted have an impact on the classification algorithm's dependability. With the aid of a Support Vector Machine, texture features are retrieved in this study using a Grey Level Co-occurrence Matrix, while form features are extracted using connected areas. Images of benign tumours, malignant tumours, and a normal brain all exhibit distinctive features. The classification of MR images can benefit from this change in feature values. A SVM classifier will receive the features that were thusly obtained for training and testing and further able to classify the abnormalities in brain images.

Evaluation of nutritional status and clinical depression classification using an explainable machine learning method.

Depression is a prevalent disorder worldwide, with potentially severe implications. It contributes significantly to an increased risk of diseases associated with multiple risk factors. Early accurate diagnosis of depressive symptoms is a critical first step toward management, intervention, and prevention. Various nutritional and dietary compounds have been suggested to be involved in the onset, maintenance, and severity of depressive disorders. Despite the challenges to better understanding the association between nutritional risk factors and the occurrence of depression, assessing the interplay of these markers through supervised machine learning remains to be fully explored. This study aimed to determine the ability of machine learning-based decision support methods to identify the presence of depression using publicly available health data from the Korean National Health and Nutrition Examination Survey. Two exploration techniques, namely, uniform manifold approximation and projection and Pearson correlation, were performed for explanatory analysis among datasets. A grid search optimization with cross-validation was performed to fine-tune the models for classifying depression with the highest accuracy. Several performance measures, including accuracy, precision, recall, F1 score, confusion matrix, areas under the precision-recall and receiver operating characteristic curves, and calibration plot, were used to compare classifier performances. We further investigated the importance of the features provided: visualized interpretation using ELI5, partial dependence plots, and local interpretable using model-agnostic explanations and Shapley additive explanation for the prediction at both the population and individual levels. The best model achieved an accuracy of 86.18% for XGBoost and an area under the curve of 84.96% for the random forest model in original dataset and the XGBoost algorithm with an accuracy of 86.02% and an area under the curve of 85.34% in the quantile-based dataset. The explainable results revealed a complementary observation of the relative changes in feature values, and, thus, the importance of emergent depression risks could be identified. The strength of our approach is the large sample size used for training with a fine-tuned model. The machine learning-based analysis showed that the hyper-tuned model has empirically higher accuracy in classifying patients with depressive disorder, as evidenced by the set of interpretable experiments, and can be an effective solution for disease control.

Sub-regional analysis of the parotid glands: model development for predicting late xerostomia with radiomics features in head and neck cancer patients

Background The irradiation of sub-regions of the parotid has been linked to xerostomia development in patients with head and neck cancer (HNC). In this study, we compared the xerostomia classification performance of radiomics features calculated on clinically relevant and de novo sub-regions of the parotid glands of HNC patients. Material and Methods All patients (N = 117) were treated with TomoTherapy in 30–35 fractions of 2–2.167 Gy per fraction with daily mega-voltage-CT (MVCT) acquisition for image-guidance purposes. Radiomics features (N = 123) were extracted from daily MVCTs for the whole parotid gland and nine sub-regions. The changes in feature values after each complete week of treatment were considered as predictors of xerostomia (CTCAEv4.03, grade ≥ 2) at 6 and 12 months. Combinations of predictors were generated following the removal of statistically redundant information and stepwise selection. The classification performance of the logistic regression models was evaluated on train and test sets of patients using the Area Under the Curve (AUC) associated with the different sub-regions at each week of treatment and benchmarked with the performance of models solely using dose and toxicity at baseline. Results In this study, radiomics-based models predicted xerostomia better than standard clinical predictors. Models combining dose to the parotid and xerostomia scores at baseline yielded an AUCtest of 0.63 and 0.61 for xerostomia prediction at 6 and 12 months after radiotherapy while models based on radiomics features extracted from the whole parotid yielded a maximum AUCtest of 0.67 and 0.75, respectively. Overall, across sub-regions, maximum AUCtest was 0.76 and 0.80 for xerostomia prediction at 6 and 12 months. Within the first two weeks of treatment, the cranial part of the parotid systematically yielded the highest AUCtest. Conclusion Our results indicate that variations of radiomics features calculated on sub-regions of the parotid glands can lead to earlier and improved prediction of xerostomia in HNC patients.

An empirical study on how humans appreciate automated counterfactual explanations which embrace imprecise information

The explanatory capacity of interpretable fuzzy rule-based classifiers is usually limited to offering explanations for the predicted class only. A lack of potentially useful explanations for non-predicted alternatives can be overcome by designing methods for the so-called counterfactual reasoning. Nevertheless, state-of-the-art methods for counterfactual explanation generation require special attention to human evaluation aspects, as the final decision upon the classification under consideration is left for the end user. In this paper, we first introduce novel methods for qualitative and quantitative counterfactual explanation generation. Then, we carry out a comparative analysis of qualitative explanation generation methods operating on (combinations of) linguistic terms as well as a quantitative method suggesting precise changes in feature values. Then, we propose a new metric for assessing the perceived complexity of the generated explanations. Further, we design and carry out two human evaluation experiments to assess the explanatory power of the aforementioned methods. As a major result, we show that the estimated explanation complexity correlates well with the informativeness, relevance, and readability of explanations perceived by the targeted study participants. This fact opens the door to using the new automatic complexity metric for guiding multi-objective evolutionary explainable fuzzy modeling in the near future.

Abnormal behavior detection of stationary objects in surveillance videos with visualization and classification

SUMMARYAnomaly detection in video systems has been popular over several years. It is still challenging to detect anomalies in a static object. To manage this objective, we focus on changes in the position of a stationary object in videos. In a normal scenario, the pixel values of the static object are fixed while in abnormal motion the fixed values change. We introduce a new concept to determine anomalies based on manual annotations in each video frame, only over a part of a static object in a frame such that it can be taken as a reference for the whole. Through color channel splitting we determine mask image, from which handcrafted features such as scratch area, perimeter, equivalent diameter and density are calculated. In the next step, we analyze frame‐wise changes in feature values using a linear regression model, feature values are constant when the object remains stationary while there is a rise or fall in values when an object changes location. We classify feature values through anomaly scores and thresholds. In this model, we are evaluating our proposed framework on 12 real‐time video datasets. Results are compared with existing techniques which are outperforming in terms of accuracy, mean square error and area under the curve.

Investigating the effect of sound in horror clip on the cardiac electrophysiology of young adults using wavelet packet decomposition and machine learning classifiers

The effect of horror sound on electro-cardiac activity is not yet sufficiently explored despite recommendable research on different genres of music and sound. Detecting the impact of these stimuli is a complex task. The current study intends to automatically detect the horror-sound induced cardiac changes using wavelet coefficient-based features extracted from the electrocardiograph signals and several machine learning models. The statistical comparison of these features in the two data groups suggests a significant change in different feature values after exposure to horror sound while watching a horror clip. The statistically discriminative features were employed as input to nine machine learning (ML) models, namely Naïve Bayes, Fast large margin, Logistic regression, generalized linear model, Decision tree, Random Forest, deep learning, gradient boosted tree, and Support vector machine. The performance of these ML models was compared using different performance measures, including accuracy, precision, F-measure, sensitivity, specificity, and area under the curve (AUC). The results suggest the gradient boosted tree is the best performing model in differentiating the horror sound-induced changes in the ECG signals.

Extrinsic parameter's adjustment and potential implications in Plasmodium falciparum malaria diagnosis.

Malaria is a major public health concern, affecting over 3.2 billion people in 91 countries. The advent of digital microscopy and Machine learning with the aim of automating Plasmodium falciparum diagnosis extensively depends on the extracted image features. The color of the cells, plasma, and stained artifacts influence the topological, geometrical, and statistical parameters being used to extract image features. During microscopic image acquisition, custom adjustments to the condenser and color temperature controls often have an influence on the extracted statistical features. But, our human visual system sub-consciously adjusts the color and retains the originality in a different lighting environment. Despite the use of appropriate image preprocessing, findings from the literature indicate that statistical feature variations exist, allowing the risk of P. falciparum misinterpretation. In order to eliminate this pervasive variation, the current work focuses on preprocessing the extracted statistical features rather than the prepossessing of the source image. It begins with the augmentation of series images for a microscopic field by inducing illumination variations during the microscopic image acquisition stage. A set of such image series is analyzed using a Nonlinear Regression Model to generalize the relationship between microscopic images acquired with variable ambient brightness and a specific feature. The projection point of the centroid feature onto the brightness parameter is identified in the model and it is denoted as the optimum brightness factor (OBF). Using the model, the feature correction factor (CF) is calculated from the rate of change of feature values over the interval OBF, and the brightness of the test image is processed. The present work has investigated OBF for selected image textural features, namely Contrast, Homogeneity, Entropy, Energy, and Correlation individually from its co-occurrence matrices. For performance analysis, the best state-of-the-art method uses selected texture as a subset feature to evaluate the effectiveness of P. falciparum malaria classification. Then, the impact of proposed feature processing is evaluated on 274 blood smear images with and without Feature Correction (FC). As a result, the "p" value is less than .05, which leads to the result that it is highly significant and the classification accuracy and F-score of P. falciparum malaria are increased.

Variability in the atrial flutter vectorcardiographic loops and non-invasive localization of circuits

Localization of atrial flutter circuits is tedious. Knowledge of this information beforehand will aid clinicians assess and plan ablation operations in advance to improve efficacy. In this article, we develop a novel classifier to determine flutter circuit localization by exploiting the variability contained in beat-to-beat series of vectorcardiographic flutter loop parameters, related to the variability of the activation circuit. Resulting classifier performance is satisfactory (maximum accuracy 0.93 for sample size of 56, sensitivity and specificity of 0.87 and 1.00 using logistic regression). Using an original method for selection of the most relevant features, it is shown that right and left AFL variability is different (right AFL has larger variability compared to left AFL) regarding higher-order statistics. On the other hand, it is hypothesized that respiratory motion alongside physical location of the two atria introduces variability in different amounts in right and left flutter circuits. This constitutes a confounding source of variability and may explain differences between right and left variability. We verify this hypothesis by removing this variability related to respiratory motion and analyzing the changes in classifier performance and variability feature values. It was concluded that respiratory motion is not the origin of discriminatory variability, and that this variability possibly originates from the AFL itself.

Harmonization of radiomic feature variability resulting from differences in CT image acquisition and reconstruction: assessment in a cadaveric liver

Studies investigating the effects of computed tomography (CT) image acquisition and reconstruction parameters have mostly been limited to non-human phantoms to limit exposure to patients. This study investigates these variations using a cadaveric liver and determines harmonization methods to mitigate these variations. A reference CT scan of a cadaveric liver was acquired along with 16 modified scans. Modified scans were obtained with altered image acquisition and reconstruction parameters. In each slice, the liver was segmented and used to calculate 142 features. Student’s t-tests assessed differences between reference and modified scans for each feature after correcting for multiple comparisons. Features were harmonized between reference and modified scans using histogram normalization, pixel resampling, Butterworth filtering, resampling and filtering combined, and ComBat harmonization. The number of features reflecting significant differences before and after harmonization were compared across imaging parameters. Reducing the field-of-view (FOV) and using coronal instead of axial scans resulted in the greatest number of features reflecting significant differences (67.6%, and 35.9%, respectively) and resulted in the greatest median relative change in feature values (25.4% and 18.2%, respectively). Changes in tube voltage, pitch, and slice interval resulted in the smallest number of features reflecting significance (0.7%) with median relative changes in feature <2%. Histogram normalization reduced or maintained the number of significantly different features for all scans, while ComBat reduced the number of significantly different features to zero for all scans. The remaining harmonization methods had mixed effects: resampling reduced the number of features reflecting significant differences for half of the imaging parameters, while filtering alone and filtering combined with resampling both reduced the number of features reflecting significance for 10 of the 16 parameters. The dependence of radiomic features on image acquisition and reconstruction parameters varies in a cadaveric liver; however, various harmonization methods have shown promise in mitigating these dependencies, particularly ComBat.

Critical evaluation and comparison of psychoacoustics, acoustics and vibration features for gear fault correlation and classification

Gear fault diagnosis has gained importance in the last few decades with the focus of fault diagnosis function for maintenance purpose. This paper investigates the ability of various vibration, acoustics and psychoacoustic features to correlate and classify faults in the gearbox. This study is motivated by the process of gearbox fault diagnosis at the end of the assembly line inspection of gearbox. It is observed that during the end of assembly line inspection, the gearbox is operated on a test bench and operator takes a decision about the presence of fault by listening to the sound emitted by the gearbox. This decision is based on an operator’s judgement and past experience which involves subjectivity. Efforts are made to address this issue of end of the line inspection, by applying a scientific, objective psychoacoustic based technique which works on similar principles of listening. Experiments are performed in a laboratory by simulating four types of gear faults and vibration and acoustic signals are acquired to extract statistical features of vibration and acoustic signal along with psychoacoustic features. It is observed that different features respond to faults in different ways and changes in feature values are also dependent on loading condition. Therefore, correlating these features with faults is not simple and need intelligent techniques for correlating faults to features. This correlation of faults to features is attempted with different techniques like ANN and discriminant classifier. It is found that the psychoacoustic features have better ability to classify faults compared to acoustics and vibration statistical features which are found to be 93.02% using quadratic discriminant classifier and 95.93% using multilayer feed-forward back-propagation neural network. It is shown that the psychoacoustic based fault identification technique can be applied for the end of the assembly line inspection of gearbox.

Correlating Changes in the Epithelial Gland Tissue With Advancing Colorectal Cancer Histologic Grade, Using IHC Stained for AIB1 Expression Biopsy Material

The objective of this study was to study the textural and color changes occurring in the epithelial gland tissue with advancing colorectal cancer (CRC), utilizing immunohistochemical stain for AIB1 expression biopsy material. Clinical material comprised biopsy specimens of 67 patients with a diagnosis of CRC. Two experienced pathologists used H&E-stained material for grading CRC lesions and immunohistochemical (IHC) stain for AIB1 expression. Twenty six patients were diagnosed with grade I, 28 with grade II, and 13 with grade III CRC. Guided by pathologists, we selected the regions of interest from AIB1-digitized images of each patient, encompassing the epithelial gland, and we computed 69 features, quantifying textural and color properties of the AIB1-stained lesions. We evaluated the statistical differences between grades by means of the Wilcoxon statistical test for each feature, and we assessed changes in feature values with advancing tumor grade by means of the Point Biserial Correlation. Statistical analysis revealed 14 single features, quantifying textural and color properties of the epithelial gland, which sustained statistically significant differences between LG-CRC and HG-CRC cases. These features were drawn from the gray-level image histogram, the cooccurrence matrix, the run length matrix, the discrete wavelet transform, the Tamura method, and the L*a*b color transform. A systematic statistical analysis of AIB1-stained biopsy material showed that high-grade CRC lesions contain higher intensity levels, appear coarser, are more homogeneous with smooth variation across the image, have lower contrast that is slowly varying across the image, have lower AIB1 staining, and have lower edges. A combination of textural and color attributes, evaluating image gray-tone distribution, textural roughness, inhomogeneity, AIB1 staining, and image coarseness should be considered in evaluating AIB1-stained CRC lesions.

The complexity of tumor shape, spiculatedness, correlates with tumor radiomic shape features

Radiomics extracts high-throughput quantitative data from medical images to contribute to precision medicine. Radiomic shape features have been shown to correlate with patient outcomes. However, how radiomic shape features vary in function of tumor complexity and tumor volume, as well as with method used for meshing and voxel resampling, remains unknown. The aims of this study are to create tumor models with varying degrees of complexity, or spiculatedness, and evaluate their relationship with quantitatively extracted shape features. Twenty-eight tumor models were mathematically created using spherical harmonics with the spiculatedness degree d being increased by increments of 3 (d = 11 to d = 92). Models were 3D printed with identical bases of 5 cm, imaged with a CT scanner with two different slice thicknesses, and semi-automatically delineated. Resampling of the resulting masks on a 1 × 1 × 1 mm3 grid was performed, and the voxel size of each model was then calculated to eliminate volume differences. Four MATLAB-based algorithms (isosurface (M1), isosurface filter (M2), isosurface remeshing (M3), and boundary (M4)) were used to extract nine 3D features (Volume, Surface area, Surface-to-volume, Compactness1, Compactness2, Compactness3, Spherical Disproportion, Sphericity and Fractional Concavity). To quantify the impact of 3D printing, acquisition, segmentation and meshing, features were computed directly from the stereolithography (STL) file format that was used for 3D printing, and compared to those computed. Changes in feature values between 0.6 and 2 mm slice acquisitions were also compared. Spearman’s rank-order correlation coefficients were computed to determine the relationship of each shape feature with spiculatedness for each of the four meshing algorithms. Percent changes were calculated between shape features extracted from the original and resampled contoured images to evaluate the influence of spatial resampling. Finally, the percent change in shape features when the volume was changed from 25% to 150% of their original volume was quantified for three distinct tumor models and compared to the percent change observed when modifying the spiculatedness of the model from d = 11 to d = 92. Values extracted using isosurface remeshing method are the closest to the STL reference ones, with mean differences less than 10.8% (Compactness2) for all features. Seven of the eight features had strong significant correlations with tumor model complexity irrespective of the meshing algorithm (r > 0.98, p < 10-4), with fractional concavity having the lowest correlation coefficient (r = 0.83, p < 10-4, M2). Comparisons of features extracted from the 0.6 and 2 mm slice thicknesses showed that mean differences were from 2.1% (Compactness3) to 12.7% (Compactness2) for the isosurface remeshing method. Resampling on a 1 × 1 × 1 mm3 grid resulted in between 1.3% (Compactness3) to 9.5% (Fractional Concavity) mean changes in feature values. Compactness2, Compactness3, Spherical Disproportion, Sphericity and Fractional Concavity were the features least affected by volume changes. Compactness1 had a 90.4% change with volume, which was greater than the change between the least and most spiculated models. This is the first methodological study that directly demonstrates the relationship of tumor spiculatedness with radiomic shape features, that also produced 3D tumor models, which may serve as reference phantoms for future radiomic studies. Surface Area, Surface-to-volume, and Spherical Disproportion had direct relationships with spiculatedness while the three formulas for Compactness, Sphericity and Fractional Concavity had inverse relationships. The features Compactness2, Compactness3, Spherical Disproportion, and Sphericity should be prioritized as these have minimal variations with volume changes, slice thickness and resampling.

Time stability of delta-radiomics features and the impact on patient analysis in longitudinal CT images.

This study first aims to show that the values of texture features extracted from phantoms are stable over clinical timescales. Second, that changes in patients' feature values over the course of radiation therapy (RT) are treatment induced and statistically significant. The CT datasets of a 3D printed anatomically informed texture phantom containing liver and low-contrast modules, and the homogeneous module of the Catphan 500-Series phantom, were acquired once per week over the course of a 6-week period, to simulate the timescale of conventional RT duration. A Definition AS Open CT scanner on rails (Siemens) and our institution's standard abdominal protocol were used. In each phantom module, 8 regions of interest (20cm ) were selected and 50 texture features were extracted from each module over the longitudinal dataset. The time stability of each feature was evaluated. The expected variation over the treatment timescale was quantified for each texture (module). Subsequently, the pancreas heads of 10 patients who underwent RT for adenocarcinoma of the pancreas head with a pathologic response of at least "moderate" (grade 2), were contoured on the daily CTs acquired using the same scanner. The pancreas heads were contoured on one image per week. Mean CT number, skewness, kurtosis, and coarseness were extracted from these data. The phantom modules were shown to be accurate representations of these features in the pancreas data. The change in the feature value between fractions 2 and 26 was compared with the phantom data in order to identify significant changes in feature value. Of the 50 features examined in all 3 phantom modules, 47 were found to have zero time-trend when a fit assuming homogeneous variance was used. When a fit allowing for heterogeneous variance was used, 49 features were found to have zero time-trend. Features were stable and repeatable within a feature-specific confidence interval over the 6-week period of acquisition in all three phantom modules. Changes in feature value between fractions 2 and 26 were highly patient specific. Mean CT number was found to decrease significantly in 7 of 10 patients and increase significantly in one patient. Skewness increased significantly in one patient and decreased significantly in one patient. Kurtosis decreased significantly in four patients and increased significantly in one patient. Coarseness increased significantly in seven patients and decreased significantly in one patient. Only one patient experienced no significant changes in feature value. The CT texture feature measurements of phantoms are stable and repeatable within a feature-specific confidence interval in all three phantom modules. This suggests that the changes observed in features extracted from longitudinal patient CT data may be treatment induced, and demonstrates their potentiality for early assessment of treatment response.

SU‐E‐J‐113: Effects of Deformable Registration On First‐Order Texture Maps Calculated From Thoracic Lung CT Scans

Purpose:To determine the stability of eight first‐order texture features following the deformable registration of serial computed tomography (CT) scans.Methods:CT scans at two different time points from 10 patients deemed to have no lung abnormalities by a radiologist were collected. Following lung segmentation using an in‐house program, texture maps were calculated from 32×32‐pixel regions of interest centered at every pixel in the lungs. The texture feature value of the ROI was assigned to the center pixel of the ROI in the corresponding location of the texture map. Pixels in the square ROI not contained within the segmented lung were not included in the calculation. To quantify the agreement between ROI texture features in corresponding pixels of the baseline and follow‐up texture maps, the Fraunhofer MEVIS EMPIRE10 deformable registration algorithm was used to register the baseline and follow‐up scans. Bland‐Altman analysis was used to compare registered scan pairs by computing normalized bias (nBias), defined as the feature value change normalized to the mean feature value, and normalized range of agreement (nRoA), defined as the range spanned by the 95% limits of agreement normalized to the mean feature value.Results:Each patient's scans contained between 6.8–15.4 million ROIs. All of the first‐order features investigated were found to have an nBias value less than 0.04% and an nRoA less than 19%, indicating that the variability introduced by deformable registration was low.Conclusion:The eight first‐order features investigated were found to be registration stable. Changes in CT texture maps could allow for temporal‐spatial evaluation of the evolution of lung abnormalities relating to a variety of diseases on a patient‐by‐patient basis.SGA and HA receives royalties and licensing fees through the University of Chicago for computer‐aided diagnosis technology. Research reported in this publication was supported by the National Institute Of General Medical Sciences of the National Institutes of Health under Award Number R25GM109439.

Quantitative identification of magnetic resonance imaging features of prostate cancer response following laser ablation and radical prostatectomy.

Laser interstitial thermotherapy (LITT) is a relatively new focal therapy technique for the ablation of localized prostate cancer. In this study, for the first time, we are integrating ex vivo pathology and magnetic resonance imaging (MRI) to assess the imaging characteristics of prostate cancer and treatment changes following LITT. Via a unique clinical trial, which gave us the availability of ex vivo histology and pre- and post-LITT MRIs, (1)we investigated the imaging characteristics of treatment effects and residual disease, and (2)evaluated treatment-induced feature changes in the ablated area relative to the residual disease. First, a pathologist annotated the ablated area and the residual disease on the ex vivo histology. Subsequently, we transferred the annotations to the post-LITT MRI using a semi-automatic elastic registration. The pre- and post-LITT MRIs were registered and features were extracted. A scoring metric based on the change in median pre- and post-LITT feature values was introduced, which allowed us to identify the most treatment responsive features. Our results show that (1)image characteristics for treatment effects and residual disease are different, and (2)the change of feature values between pre- and post-LITT MRIs can be a quantitative biomarker for treatment response. Finally, using feature change improved discrimination between the residual disease and treatment effects.

Lung texture in serial thoracic CT scans: correlation with radiologist-defined severity of acute changes following radiation therapya

This study examines the correlation between the radiologist-defined severity of normal tissue damage following radiation therapy (RT) for lung cancer treatment and a set of mathematical descriptors of computed tomography (CT) scan texture (‘texture features’). A pre-therapy CT scan and a post-therapy CT scan were retrospectively collected under IRB approval for each of the 25 patients who underwent definitive RT (median dose: 66 Gy). Sixty regions of interest (ROIs) were automatically identified in the non-cancerous lung tissue of each post-therapy scan. A radiologist compared post-therapy scan ROIs with pre-therapy scans and categorized each as containing no abnormality, mild abnormality, moderate abnormality, or severe abnormality. Twenty texture features that characterize gray-level intensity, region morphology, and gray-level distribution were calculated in post-therapy scan ROIs and compared with anatomically matched ROIs in the pre-therapy scan. Linear regression and receiver operating characteristic (ROC) analysis were used to compare the percent feature value change (ΔFV) between ROIs at each category of visible radiation damage. Most ROIs contained no (65%) or mild abnormality (30%). ROIs with moderate (3%) or severe (2%) abnormalities were observed in 9 patients. For 19 of 20 features, ΔFV was significantly different among severity levels. For 12 features, significant differences were observed at every level. Compared with regions with no abnormalities, ΔFV for these 12 features increased, on average, by 1.5%, 12%, and 30%, respectively, for mild, moderate, and severe abnormalitites. Area under the ROC curve was largest when comparing ΔFV in the highest severity level with the remaining three categories (mean AUC across features: 0.84). In conclusion, 19 features that characterized the severity of radiologic changes from pre-therapy scans were identified. These features may be used in future studies to quantify acute normal lung tissue damage following RT.

TU‐A‐12A‐04: Quantitative Texture Features Calculated in Lung Tissue From CT Scans Demonstrate Consistency Between Two Databases From Different Institutions

Purpose:To evaluate the consistency of computed tomography (CT) scan texture features, previously identified as stable in a healthy patient cohort, in esophageal cancer patient CT scans.Methods:116 patients receiving radiation therapy (median dose: 50.4Gy) for esophageal cancer were retrospectively identified. For each patient, diagnostic‐quality pre‐therapy (0‐183 days) and post‐therapy (5‐120 days) scans (mean voxel size: 0.8mm×0.8mm×2.5mm) and a treatment planning scan and associated dose map were collected. An average of 501 32×32‐pixel ROIs were placed randomly in the lungs of each pre‐therapy scan. ROI centers were mapped to corresponding locations in post‐therapy and planning scans using the displacement vector field output by demons deformable registration. Only ROIs with mean dose &lt;5Gy were analyzed, as these were expected to contain minimal post‐treatment damage. 140 texture features were calculated in pre‐therapy and post‐therapy scan ROIs and compared using Bland‐Altman analysis. For each feature, the mean feature value change and the distance spanned by the 95% limits of agreement were normalized to the mean feature value, yielding normalized range of agreement (nRoA) and normalized bias (nBias). Using Wilcoxon signed rank tests, nRoA and nBias were compared with values computed previously in 27 healthy patient scans (mean voxel size: 0.67mm×0.67mm×1mm) acquired at a different institution.Results:nRoA was significantly (p&lt;0.001) larger in cancer patients than healthy patients. Differences in nBias were not significant (p=0.23). The 20 features identified previously as having nRoA&lt;20% for healthy patients had the lowest nRoA values in the current database, with an average increase of 5.6%.Conclusion:Despite differences in CT scanner type, scan resolution, and patient health status, the same 20 features remained stable (i.e., low variability and bias) in the absence of disease changes for databases from two institutions. Identification of these features is the first step towards quantifying radiation‐induced changes between preand post‐therapy scans.Supported, in part, by NIH Grant Nos. S10 RR021039, and P30 CA14599, the Virginia and D. K. Ludwig Fund for Cancer Research, Imaging Research Institute, Biological Sciences Division, The University of Chicago, and The Institute for Translational Medicine Pilot Award, The University of Chicago.