Inherent bias in the artificial intelligence (AI)-model brings inaccuracies and variabilities during clinical deployment of the model. It is challenging to recognize the source of bias in AI-model due to variations in datasets and black box nature of system design. Additionally, there is no distinct process to identify the potential source of bias in the AI-model. To the best of our knowledge, this is the first review of its kind that addresses the bias in AI-model by categorizing 48 studies into three classes, namely, point-based, image-based, and hybrid-based AI-models. Selection strategy using PRISMA is adopted to select the 72 crucial AI studies for identifying bias in AI models. Using the three classes, bias is identified in these studies based on 44 critical AI attributes. Bias in the AI-models is computed by analytical, butterfly, and ranking-based bias models. These bias models were evaluated using two experts and compared using variability analysis. AI-studies that lacked sufficient AI-attributes are more prone to risk-of-bias (RoB) in all three classes. Studies with high RoB loses fins in the butterfly model. It has been analyzed that the majority of the studies in healthcare suffer from data bias and algorithmic bias due to incomplete specifications mentioned in the design protocol and weak AI design exploited for prediction.

SARS-CoV-2 has infected over ∼165 million people worldwide causing Acute Respiratory Distress Syndrome (ARDS) and has killed ∼3.4 million people. Artificial Intelligence (AI) has shown to benefit in the biomedical image such as X-ray/Computed Tomography in diagnosis of ARDS, but there are limited AI-based systematic reviews (aiSR). The purpose of this study is to understand the Risk-of-Bias (RoB) in a non-randomized AI trial for handling ARDS using novel AtheroPoint-AI-Bias (AP(ai)Bias). Our hypothesis for acceptance of a study to be in low RoB must have a mean score of 80% in a study. Using the PRISMA model, 42 best AI studies were analyzed to understand the RoB. Using the AP(ai)Bias paradigm, the top 19 studies were then chosen using the raw-cutoff of 1.9. This was obtained using the intersection of the cumulative plot of "mean score vs. study" and score distribution. Finally, these studies were benchmarked against ROBINS-I and PROBAST paradigm. Our observation showed that AP(ai)Bias, ROBINS-I, and PROBAST had only 32%, 16%, and 26% studies, respectively in low-moderate RoB (cutoff>2.5), however none of them met the RoB hypothesis. Further, the aiSR analysis recommends six primary and six secondary recommendations for the non-randomized AI for ARDS. The primary recommendations for improvement in AI-based ARDS design inclusive of (i) comorbidity, (ii) inter-and intra-observer variability studies, (iii) large data size, (iv) clinical validation, (v) granularity of COVID-19 risk, and (vi) cross-modality scientific validation. The AI is an important component for diagnosis of ARDS and the recommendations must be followed to lower the RoB.

Diabetes Mellitus, often referred to as diabetes, is a chronic disease that affects a vast majority of world population. The percentage of people affected is increasing every year. Diabetes is very difficult to cure. It can only be kept under control

Epilepsy, a neurological disorder, is characterized by the recurrence of seizures. Electroencephalogram (EEG) signals, which are used to detect the presence of seizures, are non-linear and dynamic in nature. Visual inspection of the EEG signals for detection of normal, interictal, and ictal activities is a strenuous and time-consuming task due to the huge volumes of EEG segments that have to be studied. Therefore, non-linear methods are being widely used to study EEG signals for the automatic monitoring of epileptic activities. The aim of our work is to develop a Computer Aided Diagnostic (CAD) technique with minimal pre-processing steps that can classify all the three classes of EEG segments, namely normal, interictal, and ictal, using a small number of highly discriminating non-linear features in simple classifiers. To evaluate the technique, segments of normal, interictal, and ictal EEG segments (100 segments in each class) were used. Non-linear features based on the Higher Order Spectra (HOS), two entropies, namely the Approximation Entropy (ApEn) and the Sample Entropy (SampEn), and Fractal Dimension and Hurst Exponent were extracted from the segments. Significant features were selected using the ANOVA test. After evaluating the performance of six classifiers (Decision Tree, Fuzzy Sugeno Classifier, Gaussian Mixture Model, K-Nearest Neighbor, Support Vector Machine, and Radial Basis Probabilistic Neural Network) using a combination of the selected features, we found that using a set of all the selected six features in the Fuzzy classifier resulted in 99.7% classification accuracy. We have demonstrated that our technique is capable of achieving high accuracy using a small number of features that accurately capture the subtle differences in the three different types of EEG (normal, interictal, and ictal) segments. The technique can be easily written as a software application and used by medical professionals without any extensive training and cost. Such software can evolve into an automatic seizure monitoring application in the near future and can aid the doctors in providing better and timely care for the patients suffering from epilepsy.

Electroencephalogram (EEG) signals, which record the electrical activity in the brain, are useful for assessing the mental state of a person. Since these signals are nonlinear and non-stationary in nature, it is very difficult to decipher the useful information from them using conventional statistical and frequency domain methods. Hence, the application of nonlinear time series analysis to EEG signals could be useful to study the dynamical nature and variability of the brain signals. In this paper, we propose a Computer Aided Diagnostic (CAD) technique for the automated identification of normal and alcoholic EEG signals using nonlinear features. We first extract nonlinear features such as Approximate Entropy (ApEn), Largest Lyapunov Exponent (LLE), Sample Entropy (SampEn), and four other Higher Order Spectra (HOS) features, and then use them to train Support Vector Machine (SVM) classifier of varying kernel functions: 1st, 2nd, and 3rd order polynomials and a Radial basis function (RBF) kernel. Our results indicate that these nonlinear measures are good discriminators of normal and alcoholic EEG signals. The SVM classifier with a polynomial kernel of order 1 could distinguish the two classes with an accuracy of 91.7%, sensitivity of 90% and specificity of 93.3%. As a pre-analysis step, the EEG signals were tested for nonlinearity using surrogate data analysis and we found that there was a significant difference in the LLE measure of the actual data and the surrogate data.

The aim of this paper is to describe a novel and completely automated technique for carotid artery (CA) recognition, far (distal) wall segmentation, and intima-media thickness (IMT) measurement, which is a strong clinical tool for risk assessment for cardiovascular diseases. The architecture of completely automated multiresolution edge snapper (CAMES) consists of the following two stages: 1) automated CA recognition based on a combination of scale-space and statistical classification in a multiresolution framework and 2) automated segmentation of lumen-intima (LI) and media-adventitia (MA) interfaces for the far (distal) wall and IMT measurement. Our database of 365 B-mode longitudinal carotid images is taken from four different institutions covering different ethnic backgrounds. The ground-truth (GT) database was the average manual segmentation from three clinical experts. The mean distance ± standard deviation of CAMES with respect to GT profiles for LI and MA interfaces were 0.081 ± 0.099 and 0.082 ± 0.197 mm, respectively. The IMT measurement error between CAMES and GT was 0.078 ± 0.112 mm. CAMES was benchmarked against a previously developed automated technique based on an integrated approach using feature-based extraction and classifier (CALEX). Although CAMES underestimated the IMT value, it had shown a strong improvement in segmentation errors against CALEX for LI and MA interfaces by 8% and 42%, respectively. The overall IMT measurement bias for CAMES improved by 36% against CALEX. Finally, this paper demonstrated that the figure-of-merit of CAMES was 95.8% compared with 87.4% for CALEX. The combination of multiresolution CA recognition and far-wall segmentation led to an automated, low-complexity, real-time, and accurate technique for carotid IMT measurement. Validation on a multiethnic/multi-institutional data set demonstrated the robustness of the technique, which can constitute a clinically valid IMT measurement for assistance in atherosclerosis disease management.

Over the past two decades, intravascular ultrasound (IVUS) image segmentation has remained a challenge for researchers while the use of this imaging modality is rapidly growing in catheterization procedures and in research studies. IVUS provides cross-sectional grayscale images of the arterial wall and the extent of atherosclerotic plaques with high spatial resolution in real time. In this paper, we review recently developed image processing methods for the detection of media-adventitia and luminal borders in IVUS images acquired with different transducers operating at frequencies ranging from 20 to 45 MHz. We discuss methodological challenges, lack of diversity in reported datasets, and weaknesses of quantification metrics that make IVUS segmentation still an open problem despite all efforts. In conclusion, we call for a common reference database, validation metrics, and ground-truth definition with which new and existing algorithms could be benchmarked.

The unpredictability of the occurrence of epileptic seizures makes it difficult to detect and treat this condition effectively. An automatic system that characterizes epileptic activities in EEG signals would allow patients or the people near them to take appropriate precautions, would allow clinicians to better manage the condition, and could provide more insight into these phenomena thereby revealing important clinical information. Various methods have been proposed to detect epileptic activity in EEG recordings. Because of the nonlinear and dynamic nature of EEG signals, the use of nonlinear Higher Order Spectra (HOS) features is a seemingly promising approach. This paper presents the methodology employed to extract HOS features (specifically, cumulants) from normal, interictal, and epileptic EEG segments and to use significant features in classifiers for the detection of these three classes. In this work, 300 sets of EEG data belonging to the three classes were used for feature extraction and classifier development and evaluation. The results show that the HOS based measures have unique ranges for the different classes with high confidence level (p-value < 0.0001). On evaluating several classifiers with the significant features, it was observed that the Support Vector Machine (SVM) presented a high detection accuracy of 98.5% thereby establishing the possibility of effective EEG segment classification using the proposed technique.

Tumor vascularity is an important factor that has been shown to correlate with tumor malignancy and was demonstrated as a prognostic indicator for a wide range of cancers. Three-dimensional (3-D) power Doppler ultrasound (PDUS) offers a convenient tool for investigators to inspect the signals of blood flow and vascular structures in breast cancer. In this paper, a new computer-aided diagnosis (CAD) system for quantifying Doppler ultrasound images based on 3-D thinning algorithm and neural network is proposed. We extracted the skeleton of blood vessels from 3-D PDUS data to facilitate the capturing of morphological changes. Nine features including vessel-to-volume ratio, number of vascular trees, length of vessels, number of branching, mean of radius, number of cycles, and three tortuosity measures, were extracted from the thinning result. Benign and malignant tumors can therefore be differentiated by a score computed by a multilayered perceptron (MLP) neural network using these features as parameters. The proposed system was tested on 221 breast tumors, including 110 benign and 111 malignant lesions. The accuracy, sensitivity, specificity, and positive and negative predictive values were 88.69% (196/221), 91.89% (102/111), 85.45% (94/110), 86.44% (102/118), and 91.26% (94/103), respectively. The Az value of the ROC curve was 0.94. The results demonstrate a correlation between the morphology of blood vessels and tumor malignancy, indicating that the newly proposed method can retrieves a high accuracy in the classification of benign and malignant breast tumors.

Digital subtraction angiography (DSA) is an important tool in interventional procedures and enables the surgeon to visualize the blood vessels in the projection X-ray images. Due to the motion of patient as well as motion of internal tissues constituting the background anatomy of the patient, the difference images may contain motion artifacts. The artifacts due to motion may be severe enough to make the visualization useless or erroneous and the images need to be motion compensated. Image registration is used for motion correction of DSA images such that the background mask image is placed in coordinates of the blood vessel enhanced image. The background structures are aligned as a result of image registration and are therefore removed from the subtraction image. There exist a large number of image registration techniques depending upon the application and the available information. In this paper, we compare two intensity based image registration techniques for motion correction of DSA images using signal-to-noise ratio as the evaluation metric. The methods discussed are: inverse consistent linear elastic image registration using B-splines and modified demons method, using a hierarchical strategy that focuses on region of intensity differences, like HAMMER. Both methods derive the driving function from image intensities, but impose different kind of constraints.

Breast skin-line extraction in mammograms is useful to radiologists and image processing scientists as it aids in the analysis of mammograms. Prior detection and delineation of the skin-line can reduce the effects of background noise and artifacts on procedures for image enhancement and detection of signs of breast cancer. The proposed system works in the following way. An initial estimate of the skin-line is computed using a combination of adaptive thresholding (T. Ojala et al., 2001) and connected-component analysis. The novelty of our dependency approach for skin-line estimation lies in the way we compute the Euclidean distance constraints between the initial skin-line and the stroma edge computed via bimodal histogram analysis. Because the Euclidean distance from the edge of the stroma to the actual skin-line is usually uniform, these constraints are propagated to estimate the upper or lower skin-line portions. The selection of the constrained region is based on a greedy algorithm, which is also a new component in our system. We evaluated the performance of our skin-line estimation algorithm by comparing the estimated boundary with respect to the ground-truth boundary drawn by an expert radiologist. We used polyline distance metrics for error measurement (Y. Sun et al, 2005). As part of our protocol, we compared our dependency approach methodology with a deformable model strategy (see Ferrari et al. (2004)). On a dataset of 82 images from the MIAS database (J. Suckling et al., 1994), using our dependency approach, the polyline distance error metric yielded a mean error of 3.28 pixels with a standard deviation of 2.17 pixels. In comparison, the deformable model strategy (R.J. Ferrari et al., 2004) yielded a mean error of 4.92 pixels and a standard deviation of 1.91 pixels. The results obtained have been verified by radiologists, who have indicated that the improvement obtained is clinically significant.

In computer-aided analysis of mammograms, a fast processing analysis for mammograms benefits and facilitate the real time application and is also useful for radiologist to do on-line diagnosis. Down sampling is widely applied to reduce the size of large images and improve the processing speed. This paper presents the performance evaluations among several interpolation techniques (bilinear, bicubic, wavelet and Lanczos) for ultrasound breast phantom data. We also compared lesion segmentation results of down sampled images with the results of original images. Two major metrics: the Hausdorff distance measure (HDM) and polyline distance measure (PDM) were applied to measure the performance of the segmentation. We conclude that without scarifying the quality of ultrasound images, we increase the speed of system processing using our down sampling strategies. Moreover, among the four different technologies, the Lanczos algorithm can achieve the best performance.

This paper presents a use of geometric-based method integrated with classifier for lumen wall estimation using MR plaque volumes. The following are the new things the readers will observe when it comes to plaque imaging. (a) Application of three different sets of : classifiers (fuzzy, Markovian and graph-based) for lumen region classification in plaque MR volumes. These classifiers are used in multi-resolution framework. (b) Usage of rule-based region merging applied to the sub-classes of lumen region. (c) Rotational effect on region of interest in arterial bifurcation zones for accurate lumen region identification and boundary estimation. We have used our diagnostic system with three different classifying methods on actual patient data. We measure performance of the system by computing the mean distance error with respect to boundaries traced manually by human experts. Overall, the system consists of 22,500 boundary points. The in-plane pixel resolution is 0.25 millimeters. Using Markovian classifier method, the average error was 0.61 pixels; using fuzzy classifier method, the average error was 0.62 pixels; using graph-based classifier method, the average error was 0.74 pixels. All these methods lead to error less than 0.185 mm. We also validated our system by simulating the lumen images with additive Gaussian perturbations. This system works on a Linux platform and is written in C++.

Prefiltering is a critical step in three-dimensional (3-D) segmentation of the blood vessel and its display (see the recent book by Suri et al.). This paper presents a scale-space approach for filtering the white blood and black blood angiographic volumes and its implementation issues. The raw MR angiographic volume is first converted to isotropic volume followed by 3-D higher order separable Gaussian derivative convolution with known scales to generate edge volume. The edge volume is then run by the directional processor at each voxel where the eigenvalues of the 3-D ellipsoid are computed. The vessel score per voxel is then estimated based on these three eigenvalues which suppress the nonvasculature and background structures yielding the filtered volume. The filtered volume is ray-cast to generate the maximum intensity projection images for display. The performance of the system is evaluated by computing the mean, variance, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) images. The system is run over 20 patient studies from different areas of the body such as the brain, abdomen, kidney, knee, and ankle. The computer program takes around 150 s of processing time per study for a data size of 512 x 512 x 194, which includes the complete performance evaluation. We also compare our strategy with the recently published MR filtering algorithms by Alexander et al. and Sun et al.

Vascular segmentation has recently been given much attention. This review paper has two parts. Part I focuses on the physics of magnetic resonance angiography (MRA) generation and prefiltering techniques applied to MRA data sets. Part II of the review focuses on the vessel segmentation algorithms. The first section of this paper introduces the five different sets of receive coils used with the MRI system for magnetic resonance angiography data acquisition. This section then presents the five different types of the most popular data acquisition techniques: time-of-flight (TOF), phase-contrast, contrast-enhanced, black-blood, T2-weighted, and T2*-weighted, along with their pros and cons. Section II of this paper focuses on prefiltering algorithms for MRA data sets. This is necessary for removing the background nonvascular structures in the MRA data sets. Finally, the paper concludes with a clinical discussion on the challenges and the future of the data acquisition and the automated filtering algorithms.

Vascular segmentation has recently been given much attention. This review paper has two parts. Part I of this review focused on the physics of magnetic resonance angiography (MRA) and prefiltering techniques applied to MRA. Part II of this review presents the state-of-the-art overview, status, and new achievements in vessel segmentation algorithms from MRA. The first part of this review paper is focused on the nonskeleton or direct-based techniques. Here, we present eight different techniques along with their mathematical foundations, algorithms and their pros and cons. We will also focus on the skeleton or indirect-based techniques. We will discuss three different techniques along with their mathematical foundations, algorithms and their pros and cons. This paper also includes a clinical discussion on skeleton versus nonskeleton-based segmentation techniques. Finally, we shall conclude this paper with the possible challenges, the future, and a brief summary on vascular segmentation techniques.

The class of geometric deformable models, also known as level sets, has brought tremendous impact to medical imagery due to its capability of topology preservation and fast shape recovery. In an effort to facilitate a clear and full understanding of these powerful state-of-the-art applied mathematical tools, this paper is an attempt to explore these geometric methods, their implementations and integration of regularizers to improve the robustness of these topologically independent propagating curves/surfaces. This paper first presents the origination of level sets, followed by the taxonomy of level sets. We then derive the fundamental equation of curve/surface evolution and zero-level curves/surfaces. The paper then focuses on the first core class of level sets, known as "level sets without regularizers." This class presents five prototypes: gradient, edge, area-minimization, curvature-dependent and application driven. The next section is devoted to second core class of level sets, known as "level sets with regularizers." In this class, we present four kinds: clustering-based, Bayesian bidirectional classifier-based, shape-based and coupled constrained-based. An entire section is dedicated to optimization and quantification techniques for shape recovery when used in the level set framework. Finally, the paper concludes with 22 general merits and four demerits on level sets and the future of level sets in medical image segmentation. We present applications of level sets to complex shapes like the human cortex acquired via MRI for neurological image analysis.

Given the pre- and post-contrast gadolinium magnetic resonance (MR) images of any body organ, the uptake curve estimation has diagnostic utility in the area of medicine. The rate of absorption of the contrast agent (gadolinium) by lesions can show what type of lesion it is and can also be an indication of the malignancy stage. The differences in contrast enhancement rates can be used to distinguish between benign and malignant lesions. Oncologists/radiologists and internists can classify the type of malignancy by looking at the quantitative characteristics of the tissue signal enhancement. The slopes of the uptake curve can indicate the level and type of malignancy. Such a classification is called functional segmentation. This paper presents the user-friendly MR Breast Perfusion Analysis System (BPAS), based on Motif using C/C++ and X Windows libraries, that runs on Digital Unix and XP1000 workstations supporting the Unix and Linux operating systems, respectively. We tested our software on 20 patient studies from the data collected from two major sites in the USA and Europe.

Due to the recent growth of level sets and partial differential equation (PDE) based approaches, the importance of designing regularizers has risen steadily. This paper presents a classification tree for regularizers and the design of regularization forces for the robust segmentation of static and motion imagery.

Left ventricle (LV) boundaries estimated using automated pixel-based classifiers fall short in the apical zone and are not close the ground truth boundaries as delineated by the cardiologists. These errors are not random. They have a systematic positional and orientational bias, the boundary being under-estimated in the apex zone. They have a mean apex error of about 10.5 mm. The main reasons for bias errors are the poor contrast in the apex zone, nonhomogeneous mixing of the dye with the blood inside the left ventricle and interference of the diaphragm with inferior wall. Suri et al. (1996, 1997) showed, in earlier work, techniques to remove these bias errors using calibration schemes utilizing convex information of the left ventricle that consists of the aortic valve and the apex. This convex information was linear in nature. This paper presents an improved method of two calibration schemes using the quadratic nature of convex information. They are the quadratic identical coefficient and the quadratic independent coefficient methods. Using linear calibration, the independent coefficient method yields an error of 3.47 mm without utilizing apex, while quadratic calibration with convex information has a mean error of 2.49 mm. The goal set by the cardiologists is 2.5 mm.

For complete automatic left ventricle border detection in a cardiac frame, the apex needs to be located. As the apex zone has less contrast and is harder to identify in the gray scale left ventriculograms, we use the left ventricle's longitudinal axis to assist in apex location. To automatically find the longitudinal axis of the left ventricle in any frame, we find the longest segment from either the anterior aspect of the aortic valve or the inferior aspect of the aortic valve to the left ventricle border. We assume that the ruled surface generated by the sequence of longitudinal axes through the cardiac cycle is of sufficiently simple form, so that the perturbation error, especially the large errors, between the automatically measured axis and the physician defined axis can, in part, be filtered out by a robust procedure. To discriminate those automatically determined axes that might differ significantly from the physician defined ground truth, we use Huber's weight function in an iterative reweighted least square robust fitting. The effects of inlier and outlier noise are discussed. We demonstrate that for 90% of 1200 frames of clinical data, the automatically determined apex location is less than an arc length distance of 1% of the ventricle border length from the ground truth apex location as delineated by the cardiologist.

Poor contrast in the apex zone and nonhomogeneous mixing of the dye with the blood in the left ventricle causes the left ventricle pixel-based classifiers operating on ventriculograms to yield boundaries which are not close to ground truth boundaries as delineated by the cardiologist. They have a mean boundary error of about 6.4 mm and an error of about 12.5 mm in the apex zone. These errors have a systematic positional and orientational bias, the boundary being under-estimated in the apex zone. This paper discusses two calibration methods: the identical coefficient and the independent coefficient to remove these systematic biases. From these methods, we constitute a combined algorithm which reduces the boundary error compared to either of the calibration methods. The algorithm, in a greedy way, computes which and how many vertices of the left ventricle boundary can be taken from the computed boundary of each method to best improve the performance. The corrected boundaries have a mean error of less than 3.5 mm with a standard deviation of 3.4 mm over the approximately 6/spl times/10/sup 4/ vertices in the data set of 291 studies. Our methodology reduces the mean boundary error by 2.9 millimeters over the boundary produced by the classifier. We also show the calibration algorithm performs better in the apex zone where the dye is unable to reach. For end-diastole, it reduces the error in the apex zone by 8.5 millimeters over the pixel-based classifier boundaries.