Interpretability methods of machine learning algorithms with applications in breast cancer diagnosis.

Early detection of breast cancer is a powerful tool towards decreasing its socioeconomic burden. Although, artificial intelligence (AI) methods have shown remarkable results towards this goal, their "black box" nature hinders their wide adoption in clinical practice. To address the need for AI guided breast cancer diagnosis, interpretability methods can be utilized. In this study, we used AI methods, i.e., Random Forests (RF), Neural Networks (NN) and Ensembles of Neural Networks (ENN), towards this goal and explained and optimized their performance through interpretability techniques, such as the Global Surrogate (GS) method, the Individual Conditional Expectation (ICE) plots and the Shapley values (SV). The Wisconsin Diagnostic Breast Cancer (WDBC) dataset of the open UCI repository was used for the training and evaluation of the AI algorithms. The best performance for breast cancer diagnosis was achieved by the proposed ENN (96.6% accuracy and 0.96 area under the ROC curve), and its predictions were explained by ICE plots, proving that its decisions were compliant with current medical knowledge and can be further utilized to gain new insights in the pathophysiological mechanisms of breast cancer. Feature selection based on features' importance according to the GS model improved the performance of the RF (leading the accuracy from 96.49% to 97.18% and the area under the ROC curve from 0.96 to 0.97) and feature selection based on features' importance according to SV improved the performance of the NN (leading the accuracy from 94.6% to 95.53% and the area under the ROC curve from 0.94 to 0.95). Compared to other approaches on the same dataset, our proposed models demonstrated state of the art performance while being interpretable.

A simplified approach for the determination of fitting constants in Oliver-Pharr method regarding biological samples.

The atomic force microscopy (AFM) nanoindentation regarding biological samples is a challenging procedure. Biological samples at the nanoscale can be considered as purely elastic materials under the condition that the indentation depth is very small and the indenter is smooth. However, the indenters that are commonly used are pyramidal and in several cases the indentation depths are big comparing to the dimensions of the tip apex. Hence, pyramidal indenters usually cause a permanent damage to the sample. In this case, the best model that can be applied for the data processing is the Oliver-Pharr model which takes into account the elastic-plastic behavior of the sample. The Oliver-Pharr model is based on the fitting of the unloading load-indentation data to a power law equation. In this paper a simplified procedure which ensures the accurate fitting of the unloading load-indentation data to the Oliver-Pharr model is presented and validated on experimental data obtained from a human glioma cell line. It should be noted that the proposed method can be also applied for the data fitting in the case of purely elastic response.

A Two-Level Food Classification System For People With Diabetes Mellitus Using Convolutional Neural Networks.

Accurate estimation of food's macronutrient content for people with Diabetes Mellitus (DM) is of great importance, as it determines postprandial insulin dosage. This paper introduces a classification system for food images that is adjusted to the nutritional needs of people with DM. A two-level image classification scheme, exploiting Convolutional Neural Networks (CNNs), is proposed, in order to classify an image in one of eight broad food categories with similar macronutrient content and then assign it to a specific food within that category. To this end, a visual dataset, namely NTUA-Food 2017, has been designed, consisting of 3248 images organized in eight broad food categories of totally 82 different foods. Moreover, a novel evaluation metric is proposed, which penalizes classification errors proportionally to the discrepancy in postprandial blood sugar levels between the actual and predicted class. The proposed system achieves 84.18% and 85.94% classification accuracy at the first and second level of classification, respectively, on the NTUA-Food 2017 dataset. The algorithm developed for the first level of classification on the NTUA-Food 2017 dataset improves classification accuracy on the benchmark Food Image Dataset (FID) to 97.08% outperforming previous approaches. The algorithm's mean error in terms of carbohydrate content estimation on the NTUA-Food 2017 dataset is less than 2 g per food serving.

Comparative assessment of glucose prediction models for patients with type 1 diabetes mellitus applying sensors for glucose and physical activity monitoring.

The present work presents the comparative assessment of four glucose prediction models for patients with type 1 diabetes mellitus (T1DM) using data from sensors monitoring blood glucose concentration. The four models are based on a feedforward neural network (FNN), a self-organizing map (SOM), a neuro-fuzzy network with wavelets as activation functions (WFNN), and a linear regression model (LRM), respectively. For the development and evaluation of the models, data from 10 patients with T1DM for a 6-day observation period have been used. The models' predictive performance is evaluated considering a 30-, 60- and 120-min prediction horizon, using both mathematical and clinical criteria. Furthermore, the addition of input data from sensors monitoring physical activity is considered and its effect on the models' predictive performance is investigated. The continuous glucose-error grid analysis indicates that the models' predictive performance benefits mainly in the hypoglycemic range when additional information related to physical activity is fed into the models. The obtained results demonstrate the superiority of SOM over FNN, WFNN, and LRM with SOM leading to better predictive performance in terms of both mathematical and clinical evaluation criteria.

Galectin-3, Carotid Plaque Vulnerability, and Potential Effects of Statin Therapy.

OBJECTIVES: Galectin-3, a member of galectines, a family of b-galactoside-specific lectins, has been reported to propagate vascular inflammation. The role of galectin-3 in carotid atherosclerosis is controversial. The aim of this study was to investigate the relationship of galectin-3 with plaque vulnerability in patients with high grade carotid stenosis. METHODS: This was a cross sectional study of patients undergoing carotid endarterectomy (CEA). Carotid plaques obtained from 78 consecutive patients (40 symptomatic [SG], 38 asymptomatic [AG]) undergoing CEA were histologically analyzed for galectin-3, macrophages (CD68) and laminin. Pre-operatively the biochemical profile and plaque echogenicity (gray-scale median, GSM) score were determined. RESULTS: There were no significant differences in clinical and demographic parameters between SG and AG(p > .05). The SG had a lower GSM score (44.21 ± 18.24 vs. 68.79 ± 28.79, p < .001) and a smaller positive stained area for galectin-3 (4.89 ± 1.60% vs. 12.01 ± 5.91%, p < .001) and laminin (0.88 ± 0.71% vs. 3.46 ± 2.12%, p < .001) than the AG. On the other hand, intra-plaque macrophage content was increased in SG (p < .001). For the whole cohort, symptomatic status was independently associated with intra-plaque contents of both galectin-3 (OR=0.634, p < .001), and GSM score (OR=0.750, p < .001). Notably, patients on long term statin treatment had elevated galectin-3 and lowered macrophage intra-plaque concentrations compared with those on short term treatment (p < .05). CONCLUSIONS: A low galectin-3 intra-plaque concentration seems to correlate with clinically and ultrasonically defined unstable human carotid plaques. Long term statin treatment may induce increase of intra-plaque galectin-3 concentration mediating plaque stabilization.

Carotid artery wall motion analysis from B-mode ultrasound using adaptive block matching: in silico evaluation and in vivo application.

Valid risk stratification for carotid atherosclerotic plaques represents a crucial public health issue toward preventing fatal cerebrovascular events. Although motion analysis (MA) provides useful information about arterial wall dynamics, the identification of motion-based risk markers remains a significant challenge. Considering that the ability of a motion estimator (ME) to handle changes in the appearance of motion targets has a major effect on accuracy in MA, we investigated the potential of adaptive block matching (ABM) MEs, which consider changes in image intensities over time. To assure the validity in MA, we optimized and evaluated the ABM MEs in the context of a specially designed in silico framework. ABM(FIRF2), which takes advantage of the periodicity characterizing the arterial wall motion, was the most effective ABM algorithm, yielding a 47% accuracy increase with respect to the conventional block matching. The in vivo application of ABM(FIRF2) revealed five potential risk markers: low movement amplitude of the normal part of the wall adjacent to the plaques in the radial (RMA(PWL)) and longitudinal (LMA(PWL)) directions, high radial motion amplitude of the plaque top surface (RMA(PTS)), and high relative movement, expressed in terms of radial strain (RSI(PL)) and longitudinal shear strain (LSSI(PL)), between plaque top and bottom surfaces. The in vivo results were reproduced by OF(LK(WLS)) and ABM(KF-K2), MEs previously proposed by the authors and with remarkable in silico performances, thereby reinforcing the clinical values of the markers and the potential of those MEs. Future in vivo studies will elucidate with confidence the full potential of the markers.

Automated detection of the carotid artery wall in longitudinal B-mode images using active contours initialized by the Hough transform.

In this paper, a fully automatic active-contour-based segmentation method is presented, for detecting the carotid artery wall in longitudinal B-mode ultrasound images. A Hough-transform-based methodology is used for the definition of the initial snake, followed by a gradient vector flow (GVF) snake deformation for the final contour detection. The GVF snake is based on the calculation of the image edge map and the calculation of GVF field which guides its deformation for the estimation of the real arterial wall boundaries. In twenty cases there was no significant difference between the automated segmentation and the manual diameter measurements. The sensitivity, specificity and accuracy were 0.97, 0.99 and 0.98, respectively, for both diastolic and systolic cases. In conclusion, the proposed methodology provides an accurate and reliable way to segment ultrasound images of the carotid artery.

Comparison of fractal dimension estimation algorithms for epileptic seizure onset detection.

Fractal dimension (FD) is a natural measure of the irregularity of a curve. In this study the performances of three waveform FD estimation algorithms (i.e. Katz's, Higuchi's and the k-nearest neighbour (k-NN) algorithm) were compared in terms of their ability to detect the onset of epileptic seizures in scalp electroencephalogram (EEG). The selection of parameters involved in FD estimation, evaluation of the accuracy of the different algorithms and assessment of their robustness in the presence of noise were performed based on synthetic signals of known FD. When applied to scalp EEG data, Katz's and Higuchi's algorithms were found to be incapable of producing consistent changes of a single type (either a drop or an increase) during seizures. On the other hand, the k-NN algorithm produced a drop, starting close to the seizure onset, in most seizures of all patients. The k-NN algorithm outperformed both Katz's and Higuchi's algorithms in terms of robustness in the presence of noise and seizure onset detection ability. The seizure detection methodology, based on the k-NN algorithm, yielded in the training data set a sensitivity of 100% with 10.10 s mean detection delay and a false positive rate of 0.27 h(-1), while the corresponding values in the testing data set were 100%, 8.82 s and 0.42 h(-1), respectively. The above detection results compare favourably to those of other seizure onset detection methodologies applied to scalp EEG in the literature. The methodology described, based on the k-NN algorithm, appears to be promising for the detection of the onset of epileptic seizures based on scalp EEG.

Automated detection of the carotid artery wall in B-mode ultrasound images using active contours initialized by the Hough Transform.

Automatic segmentation of the arterial lumen from ultrasound images is an important and often challenging task in clinical diagnosis. We previously used the Hough Transform (HT) to automatically extract circles from sequences of B-mode ultrasound images of transverse sections of the carotid artery. In this paper, an active-contour-based methodology is suggested, initialized by the HT circle, in an attempt to extend previous findings and to accurately detect the arterial wall boundary. The methodology is based on the generation of a gradient vector flow field, an approach attempting to overcome conventional active contours constraints. Contour estimation is then achieved by deforming the initial curve (circle) based on the gradient vector flow field. In ten normal subjects, the specificity and accuracy of the segmentation were on average higher than 0.98, whereas the sensitivity was higher than 0.82. The methodology was also applied to four subjects with atherosclerosis, in which sensitivity, specificity and accuracy were comparable to those of normal subjects. In conclusion, the HT-initialized active contours methodology provides a reliable tool to detect the carotid artery wall in ultrasound images and can be used in clinical practice.

A mathematical model of the mechanical deformation of the carotid artery wall and its application to clinical data.

The study of arterial wall mechanics, including the study of stresses and strains experienced by the vascular wall, is pivotal in our understanding of arterial physiology. In this paper, a mathematical model is provided describing the deformation of the arterial wall in terms of 6 parameters. Actual deformation waveforms were also obtained from the analysis of B-mode ultrasound image sequences of the carotid artery using block-matching. The mathematical model was fitted to the clinical data using nonlinear least squares to determine the 6 parameters for 6 different locations along the posterior and 6 along the anterior walls, on the interface between the lumen and the intima-media complex (L-IM). On the posterior wall, 6 locations were also investigated at the interface between the intima-media complex and the adventitia (IM-A) as well as at the adventitia-surrounding tissue (A-T) boundary. The root mean square error was low for all locations indicating a good fit of the proposed model to the clinical data. The amplitude of the deformation, expressed through parameter alpha, was significantly lower in the A-T interface compared to the other two interfaces. The time when the systolic peak occurs, expressed through parameter t1, was significantly lower in the L-IM interface compared to the other two interfaces. Preliminary findings from a small group of diseased wall locations suggested that the parameters alpha, b and t1 were significantly different than healthy cases. This probably reflects alterations of arterial wall mechanics due to disease. This study showed that the proposed mathematical model is a satisfactory representation of the mechanical deformation of the carotid artery wall in the radial direction and can provide valuable information in the understanding of the mechanical behavior of the arterial wall.

Computational models simulating electrophysiological activity in the basal ganglia.

Modeling of the basal ganglia has played a substantial role in gaining insight into the mechanisms involved in the computational processes performed by this elusive group of nuclei. Models of the basal ganglia have undergone revolutionary changes over the last twenty years due to the rapid accumulation of neuroscientific data. In this chapter, we present distinct modeling approaches that can be used to enhance our understanding of the functional dynamics of information processing within the basal ganglia, and their interactions with the rest of the brain. Specific examples of recently developed models dealing with the analysis of computational processing issues at different structural levels of the basal ganglia are discussed.

Characterization of carotid atherosclerotic plaques using frequency-based texture analysis and bootstrap.

Texture analysis of B-mode ultrasound images of carotid atheromatous plaque can be valuable for the accurate diagnosis of atherosclerosis. In this paper, two frequency-based texture analysis methods based on the Fourier Power Spectrum and the Wavelet Transform were used to characterize atheromatous plaques. B-mode ultrasound images of 10 symptomatic and 9 asymptomatic plaques were interrogated. A total of 109 texture features were estimated for each plaque. The bootstrap method was used to compare the mean values of the texture features extracted from the two groups. After bootstrapping, three features were found to be significantly different between the two types of plaques: the average value of the angular distribution corresponding to the wedge centered at 90 degrees, the standard deviation at scale 1 derived from the horizontal detail image, and the standard deviation at scale 2 derived from the horizontal detail image. It is concluded that frequency-based texture analysis in combination with a powerful statistical technique, such as bootstrapping, may provide valuable information about the plaque tissue type.

The genetic profile of a tumor as a determinant of its response to radiotherapy: a computer simulation of two different radiotherapeutic schemes.

The aim of this paper is to present comparative results of a tumor response to radiotherapy model, concerning two different fractionation schemes and the insulin-like growth factor I receptor expression. A clinical case of glioblastoma multiforme is selected. The model parameters are appropriately adjusted according to the literature. The results of the simulation procedure are three-dimensionally visualized and compared with clinical experience.

Stochastic modeling and validation of growth saturation and radiotherapeutic response of multicellular tumor spheroids.

An advanced three-dimensional (3D) Monte Carlo simulation model of both the avascular development of multicellular tumor spheroids and their response to radiation therapy is presented. The model is based upon a number of fundamental biological principles such as the transition between the cell cycle phases, the diffusion of oxygen and nutrients and the cell survival probabilities following irradiation. Predicted histological structure and tumor growth rates evaluated for the case of EMT6/Ro spheroids have been shown to be in agreement with published experimental data. Furthermore, the underlying structure of the tumor spheroid as well as its response to irradiation satisfactorily agrees with laboratory experience.

A three dimensional inverse scattering algorithm in the time-domain based on nonlinear optimization.

A new algorithm for solving the three-dimensional inverse scattering problem using time-domain data is presented. The algorithm is intended for application in cases where linearization using the Born or Rytov approximations is not possible. A functional is defined accounting for the error of the object and data equations and the unknown coefficients of the internal field and the object function are sought in order to minimize it. The algorithm projects the unknown internal field into a space of Gaussian basis functions aiming to decrease the number of the unknown coefficients. Time-windowing is utilized in order to reduce the size of the optimization problem.

Characterization of carotid atherosclerosis based on motion and texture features and clustering using fuzzy c-means.

Analysis of B-mode ultrasound images of the carotid atheromatous plaque includes the estimation of texture from static images and the estimation of motion from image sequences. The combination of these two types of information may be valuable for accurate diagnosis of vascular disease. The purpose of this paper was to study texture and motion patterns of carotid atherosclerosis and select the optimal combination of features that can characterize plaque. B-mode ultrasound images of 10 symptomatic and 9 asymptomatic plaques were interrogated. A total of 99 texture features were estimated using first-order statistics, second-order statistics, Laws texture energy and the fractal dimension. Only five texture features were significantly different between the two groups. In the same subjects, the motion of selected plaque regions was estimated using region tracking and block-matching and expressed through: a/maximal surface velocity (MSV), and b/maximal relative surface velocity (MRSV). MSV and MRSV were significantly lower in asymptomatic plaques suggesting more homogeneous motion patterns. Clustering using fuzzy c-means correctly classified 74% of plaques based on texture features only, and 79% of plaques based on motion features only. Classification performance reached 84% when a combination of motion and texture features was used.

A digital subtraction radiography scheme based on automatic multiresolution registration.

OBJECTIVES: To establish a digital subtraction radiography scheme for aligning clinical in vivo radiographs based on the implementations of an automatic geometric registration method and a contrast correction technique. METHODS: Thirty-five pairs of in vivo dental radiographs from four clinical studies were used in this work. First, each image pair was automatically aligned by applying a multiresolution registration strategy using the affine transformation followed by the implementation of the projective transformation at full resolution. Then, a contrast correction technique was applied in order to produce subtraction radiographs and fused images for further clinical evaluation. The performance of the proposed registration method was assessed against a manual method based on the projective transformation. RESULTS: The qualitative assessment of the experiments based on visual inspection has shown advantageous performance of the proposed automatic registration method against the manual method. Furthermore, the quantitative analysis showed statistical difference in terms of the root mean square (RMS) error estimated over the whole images and specific regions of interest. CONCLUSIONS: The proposed automatic geometric registration method is capable of aligning radiographs acquired with or without rigorous a priori standardization. The methodology is pixel-based and does not require the application of any segmentation process prior to alignment. The employed projective transformation provides a reliable model for registering intraoral radiographs. The implemented contrast correction technique sequentially applied provides subtraction radiographs and fused images for clinical evaluation regarding the evolution of a disease or the response to a therapeutic scheme.

A scenario for a web-based radiation treatment planning structure: A new tool for quality assurance procedure?

The desire to develop web-based platforms for remote collaboration among physicians and technologists is becoming a great challenge. In this paper we describe a web-based radiotherapy treatment planning (WBRTP) system to facilitate decentralized radiotherapy services by allowing remote treatment planning and quality assurance (QA) of treatment delivery. Significant prerequisites are digital storage of relevant data as well as efficient and reliable telecommunication system between collaborating units. The system of WBRTP includes video conferencing, display of medical images (CT scans, dose distributions etc), replication of selected data from a common database, remote treatment planning, evaluation of treatment technique and follow-up of the treated patients. Moreover the system features real-time remote operations in terms of tele-consulting like target volume delineation performed by a team of experts at different and distant units. An appraisal of its possibilities in quality assurance in radiotherapy is also discussed. As a conclusion, a WBRTP system would not only be a medium for communication between experts in oncology but mainly a tool for improving the QA in radiotherapy.

A 3D high-resolution gamma camera for radiopharmaceutical studies with small animals.

The results of studies conducted with a small field of view tomographic gamma camera based on a Position Sensitive Photomultiplier Tube are reported. The system has been used for the evaluation of radiopharmaceuticals in small animals. Phantom studies have shown a spatial resolution of 2mm in planar and 2-3mm in tomographic imaging. Imaging studies in mice have been carried out both in 2D and 3D. Conventional radiopharmaceuticals have been used and the results have been compared with images from a clinically used system.