Continuum removal for ground-based LWIR hyperspectral infrared imagery applying non-negative matrix factorization.

Continuum removal is vital in hyperspectral image analysis. It enables data to be used for any application and usually requires approximations or assumptions to be made. One of these approximations is related to the calculation of the spectra of the background's blackbody temperature. Here, we present a new method to calculate the continuum removal process. The proposed method eliminates the calculation for ground-based hyperspectral infrared imagery by applying two acquisition sets before and after using the heating source. The approach involves a laboratory experiment on a long-wave infrared (LWIR; 7.7-11.8 µm), with a LWIR-macro lens, an Infragold plate, and a heating source. To calculate the continuum removal process, the approach applies non-negative matrix factorization (NMF) to extract Rank-1 NMF, estimate the downwelling radiance, and compare it with that of other conventional methods. NMF uses gradient-descent-based rules (GD) and non-negative least-squares (NNLS) optimization algorithms to obtain Rank-1 NMF. A comparative analysis is performed with 1%-20% additive noise for all algorithms by using the spectral angle mapper and normalized cross correlation (NCC). Results reveal the promising performance of NMF-GD (average of 72.5% similarity percentage using NCC) and NMF-NNLS (average of 77.6% similarity percentage using NCC).

Automated assessment and tracking of human body thermal variations using unsupervised clustering.

The presented approach addresses a review of the overheating that occurs during radiological examinations, such as magnetic resonance imaging, and a series of thermal experiments to determine a thermally suitable fabric material that should be used for radiological gowns. Moreover, an automatic system for detecting and tracking of the thermal fluctuation is presented. It applies hue-saturated-value-based kernelled k-means clustering, which initializes and controls the points that lie on the region-of-interest (ROI) boundary. Afterward, a particle filter tracks the targeted ROI during the video sequence independently of previous locations of overheating spots. The proposed approach was tested during experiments and under conditions very similar to those used during real radiology exams. Six subjects have voluntarily participated in these experiments. To simulate the hot spots occurring during radiology, a controllable heat source was utilized near the subject's body. The results indicate promising accuracy for the proposed approach to track hot spots. Some approximations were used regarding the transmittance of the atmosphere, and emissivity of the fabric could be neglected because of the independence of the proposed approach for these parameters. The approach can track the heating spots continuously and correctly, even for moving subjects, and provides considerable robustness against motion artifact, which occurs during most medical radiology procedures.

Development of biological movement recognition by interaction between active basis model and fuzzy optical flow division.

Following the study on computational neuroscience through functional magnetic resonance imaging claimed that human action recognition in the brain of mammalian pursues two separated streams, that is, dorsal and ventral streams. It follows up by two pathways in the bioinspired model, which are specialized for motion and form information analysis (Giese and Poggio 2003). Active basis model is used to form information which is different from orientations and scales of Gabor wavelets to form a dictionary regarding object recognition (human). Also biologically movement optic-flow patterns utilized. As motion information guides share sketch algorithm in form pathway for adjustment plus it helps to prevent wrong recognition. A synergetic neural network is utilized to generate prototype templates, representing general characteristic form of every class. Having predefined templates, classifying performs based on multitemplate matching. As every human action has one action prototype, there are some overlapping and consistency among these templates. Using fuzzy optical flow division scoring can prevent motivation for misrecognition. We successfully apply proposed model on the human action video obtained from KTH human action database. Proposed approach follows the interaction between dorsal and ventral processing streams in the original model of the biological movement recognition. The attained results indicate promising outcome and improvement in robustness using proposed approach.

Comparative study on interaction of form and motion processing streams by applying two different classifiers in mechanism for recognition of biological movement.

Research on psychophysics, neurophysiology, and functional imaging shows particular representation of biological movements which contains two pathways. The visual perception of biological movements formed through the visual system called dorsal and ventral processing streams. Ventral processing stream is associated with the form information extraction; on the other hand, dorsal processing stream provides motion information. Active basic model (ABM) as hierarchical representation of the human object had revealed novelty in form pathway due to applying Gabor based supervised object recognition method. It creates more biological plausibility along with similarity with original model. Fuzzy inference system is used for motion pattern information in motion pathway creating more robustness in recognition process. Besides, interaction of these paths is intriguing and many studies in various fields considered it. Here, the interaction of the pathways to get more appropriated results has been investigated. Extreme learning machine (ELM) has been implied for classification unit of this model, due to having the main properties of artificial neural networks, but crosses from the difficulty of training time substantially diminished in it. Here, there will be a comparison between two different configurations, interactions using synergetic neural network and ELM, in terms of accuracy and compatibility.

Dual-Intended Deep Learning Model for Breast Cancer Diagnosis in Ultrasound Imaging.

Automated medical data analysis demonstrated a significant role in modern medicine, and cancer diagnosis/prognosis to achieve highly reliable and generalizable systems. In this study, an automated breast cancer screening method in ultrasound imaging is proposed. A convolutional deep autoencoder model is presented for simultaneous segmentation and radiomic extraction. The model segments the breast lesions while concurrently extracting radiomic features. With our deep model, we perform breast lesion segmentation, which is linked to low-dimensional deep-radiomic extraction (four features). Similarly, we used high dimensional conventional imaging throughputs and applied spectral embedding techniques to reduce its size from 354 to 12 radiomics. A total of 780 ultrasound images-437 benign, 210, malignant, and 133 normal-were used to train and validate the models in this study. To diagnose malignant lesions, we have performed training, hyperparameter tuning, cross-validation, and testing with a random forest model. This resulted in a binary classification accuracy of 78.5% (65.1-84.1%) for the maximal (full multivariate) cross-validated model for a combination of radiomic groups.

Generalized ComBat harmonization methods for radiomic features with multi-modal distributions and multiple batch effects.

Radiomic features have a wide range of clinical applications, but variability due to image acquisition factors can affect their performance. The harmonization tool ComBat is a promising solution but is limited by inability to harmonize multimodal distributions, unknown imaging parameters, and multiple imaging parameters. In this study, we propose two methods for addressing these limitations. We propose a sequential method that allows for harmonization of radiomic features by multiple imaging parameters (Nested ComBat). We also employ a Gaussian Mixture Model (GMM)-based method (GMM ComBat) where scans are split into groupings based on the shape of the distribution used for harmonization as a batch effect and subsequent harmonization by a known imaging parameter. These two methods were evaluated on features extracted with CapTK and PyRadiomics from two public lung computed tomography datasets. We found that Nested ComBat exhibited similar performance to standard ComBat in reducing the percentage of features with statistically significant differences in distribution attributable to imaging parameters. GMM ComBat improved harmonization performance over standard ComBat (- 11%, - 10% for Lung3/CAPTK, Lung3/PyRadiomics harmonizing by kernel resolution). Features harmonized with a variant of the Nested method and the GMM split method demonstrated similar c-statistics and Kaplan-Meier curves when used in survival analyses.

Improved generalized ComBat methods for harmonization of radiomic features.

Radiomic approaches in precision medicine are promising, but variation associated with image acquisition factors can result in severe biases and low generalizability. Multicenter datasets used in these studies are often heterogeneous in multiple imaging parameters and/or have missing information, resulting in multimodal radiomic feature distributions. ComBat is a promising harmonization tool, but it only harmonizes by single/known variables and assumes standardized input data are normally distributed. We propose a procedure that sequentially harmonizes for multiple batch effects in an optimized order, called OPNested ComBat. Furthermore, we propose to address bimodality by employing a Gaussian Mixture Model (GMM) grouping considered as either a batch variable (OPNested + GMM) or as a protected clinical covariate (OPNested - GMM). Methods were evaluated on features extracted with CapTK and PyRadiomics from two public lung computed tomography (CT) datasets. We found that OPNested ComBat improved harmonization performance over standard ComBat. OPNested + GMM ComBat exhibited the best harmonization performance but the lowest predictive performance, while OPNested - GMM ComBat showed poorer harmonization performance, but the highest predictive performance. Our findings emphasize that improved harmonization performance is no guarantee of improved predictive performance, and that these methods show promise for superior standardization of datasets heterogeneous in multiple or unknown imaging parameters and greater generalizability.

Impartially Validated Multiple Deep-Chain Models to Detect COVID-19 in Chest X-ray Using Latent Space Radiomics.

The COVID-19 pandemic continues to spread globally at a rapid pace, and its rapid detection remains a challenge due to its rapid infectivity and limited testing availability. One of the simply available imaging modalities in clinical routine involves chest X-ray (CXR), which is often used for diagnostic purposes. Here, we proposed a computer-aided detection of COVID-19 in CXR imaging using deep and conventional radiomic features. First, we used a 2D U-Net model to segment the lung lobes. Then, we extracted deep latent space radiomics by applying deep convolutional autoencoder (ConvAE) with internal dense layers to extract low-dimensional deep radiomics. We used Johnson-Lindenstrauss (JL) lemma, Laplacian scoring (LS), and principal component analysis (PCA) to reduce dimensionality in conventional radiomics. The generated low-dimensional deep and conventional radiomics were integrated to classify COVID-19 from pneumonia and healthy patients. We used 704 CXR images for training the entire model (i.e., U-Net, ConvAE, and feature selection in conventional radiomics). Afterward, we independently validated the whole system using a study cohort of 1597 cases. We trained and tested a random forest model for detecting COVID-19 cases through multivariate binary-class and multiclass classification. The maximal (full multivariate) model using a combination of the two radiomic groups yields performance in classification cross-validated accuracy of 72.6% (69.4-74.4%) for multiclass and 89.6% (88.4-90.7%) for binary-class classification.

Impact of Interobserver Variability in Manual Segmentation of Non-Small Cell Lung Cancer (NSCLC) Applying Low-Rank Radiomic Representation on Computed Tomography.

This study tackles interobserver variability with respect to specialty training in manual segmentation of non-small cell lung cancer (NSCLC). Four readers included for segmentation are: a data scientist (BY), a medical student (LS), a radiology trainee (MH), and a specialty-trained radiologist (SK) for a total of 293 patients from two publicly available databases. Sørensen-Dice (SD) coefficients and low rank Pearson correlation coefficients (CC) of 429 radiomics were calculated to assess interobserver variability. Cox proportional hazard (CPH) models and Kaplan-Meier (KM) curves of overall survival (OS) prediction for each dataset were also generated. SD and CC for segmentations demonstrated high similarities, yielding, SD: 0.79 and CC: 0.92 (BY-SK), SD: 0.81 and CC: 0.83 (LS-SK), and SD: 0.84 and CC: 0.91 (MH-SK) in average for both databases, respectively. OS through the maximal CPH model for the two datasets yielded c-statistics of 0.7 (95% CI) and 0.69 (95% CI), while adding radiomic and clinical variables (sex, stage/morphological status, and histology) together. KM curves also showed significant discrimination between high- and low-risk patients (p-value < 0.005). This supports that readers' level of training and clinical experience may not significantly influence the ability to extract accurate radiomic features for NSCLC on CT. This potentially allows flexibility in the training required to produce robust prognostic imaging biomarkers for potential clinical translation.

Combining radiomic phenotypes of non-small cell lung cancer with liquid biopsy data may improve prediction of response to EGFR inhibitors.

Among non-small cell lung cancer (NSCLC) patients with therapeutically targetable tumor mutations in epidermal growth factor receptor (EGFR), not all patients respond to targeted therapy. Combining circulating-tumor DNA (ctDNA), clinical variables, and radiomic phenotypes may improve prediction of EGFR-targeted therapy outcomes for NSCLC. This single-center retrospective study included 40 EGFR-mutant advanced NSCLC patients treated with EGFR-targeted therapy. ctDNA data included number of mutations and detection of EGFR T790M. Clinical data included age, smoking status, and ECOG performance status. Baseline chest CT scans were analyzed to extract 429 radiomic features from each primary tumor. Unsupervised hierarchical clustering was used to group tumors into phenotypes. Kaplan-Meier (K-M) curves and Cox proportional hazards regression were modeled for progression-free survival (PFS) and overall survival (OS). Likelihood ratio test (LRT) was used to compare fit between models. Among 40 patients (73% women, median age 62 years), consensus clustering identified two radiomic phenotypes. For PFS, the model combining radiomic phenotypes with ctDNA and clinical variables had c-statistic of 0.77 and a better fit (LRT p = 0.01) than the model with clinical and ctDNA variables alone with a c-statistic of 0.73. For OS, adding radiomic phenotypes resulted in c-statistic of 0.83 versus 0.80 when using clinical and ctDNA variables (LRT p = 0.08). Both models showed separation of K-M curves dichotomized by median prognostic score (p < 0.005). Combining radiomic phenotypes, ctDNA, and clinical variables may enhance precision oncology approaches to managing advanced non-small cell lung cancer with EGFR mutations.

Interactive Machine Learning-Based Multi-Label Segmentation of Solid Tumors and Organs.

We seek the development and evaluation of a fast, accurate, and consistent method for general-purpose segmentation, based on interactive machine learning (IML). To validate our method, we identified retrospective cohorts of 20 brain, 50 breast, and 50 lung cancer patients, as well as 20 spleen scans, with corresponding ground truth annotations. Utilizing very brief user training annotations and the adaptive geodesic distance transform, an ensemble of SVMs is trained, providing a patient-specific model applied to the whole image. Two experts segmented each cohort twice with our method and twice manually. The IML method was faster than manual annotation by 53.1% on average. We found significant (p < 0.001) overlap difference for spleen (DiceIML/DiceManual = 0.91/0.87), breast tumors (DiceIML/DiceManual = 0.84/0.82), and lung nodules (DiceIML/DiceManual = 0.78/0.83). For intra-rater consistency, a significant (p = 0.003) difference was found for spleen (DiceIML/DiceManual = 0.91/0.89). For inter-rater consistency, significant (p < 0.045) differences were found for spleen (DiceIML/DiceManual = 0.91/0.87), breast (DiceIML/DiceManual = 0.86/0.81), lung (DiceIML/DiceManual = 0.85/0.89), the non-enhancing (DiceIML/DiceManual = 0.79/0.67) and the enhancing (DiceIML/DiceManual = 0.79/0.84) brain tumor sub-regions, which, in aggregation, favored our method. Quantitative evaluation for speed, spatial overlap, and consistency, reveals the benefits of our proposed method when compared with manual annotation, for several clinically relevant problems. We publicly release our implementation through CaPTk (Cancer Imaging Phenomics Toolkit) and as an MITK plugin.

Detecting Vasodilation as Potential Diagnostic Biomarker in Breast Cancer Using Deep Learning-Driven Thermomics.

Breast cancer is the most common cancer in women. Early diagnosis improves outcome and survival, which is the cornerstone of breast cancer treatment. Thermography has been utilized as a complementary diagnostic technique in breast cancer detection. Artificial intelligence (AI) has the capacity to capture and analyze the entire concealed information in thermography. In this study, we propose a method to potentially detect the immunohistochemical response to breast cancer by finding thermal heterogeneous patterns in the targeted area. In this study for breast cancer screening 208 subjects participated and normal and abnormal (diagnosed by mammography or clinical diagnosis) conditions were analyzed. High-dimensional deep thermomic features were extracted from the ResNet-50 pre-trained model from low-rank thermal matrix approximation using sparse principal component analysis. Then, a sparse deep autoencoder designed and trained for such data decreases the dimensionality to 16 latent space thermomic features. A random forest model was used to classify the participants. The proposed method preserves thermal heterogeneity, which leads to successful classification between normal and abnormal subjects with an accuracy of 78.16% (73.3-81.07%). By non-invasively capturing a thermal map of the entire tumor, the proposed method can assist in screening and diagnosing this malignancy. These thermal signatures may preoperatively stratify the patients for personalized treatment planning and potentially monitor the patients during treatment.