Abstract The computational electroencephalogram (EEG) is recently garnering significant attention in examining whether the quantitative EEG (qEEG) features can be used as new predictors for the prediction of recovery in moderate traumatic brain injury (TBI). However, the brain’s recorded electrical activity has always been contaminated with artifacts, which in turn further impede the subsequent processing steps. As a result, it is crucial to devise a strategy for meticulously flagging and extracting clean EEG data to retrieve high-quality discriminative features for successful model development. This work proposed the use of multiple artifact rejection algorithms (MARA), which is an independent component analysis (ICA)-based algorithm, to eliminate artifacts automatically, and explored their effects on the predictive performance of the random undersampling boosting (RUSBoost) model. Continuous EEG were acquired using 64 electrodes from 27 moderate TBI patients at four weeks to one-year post-accident. The MARA incorporates an artifact removal stage based on ICA prior to RUSBoost, SVM, DT, and k-NN classification. The area under the curve (AUC) of RUS Boost was higher in absolute power spectral density (PSD) in AUCδ = 0.75, AUCα = 0.73 and AUCθ = 0.71 bands than SVM, DT, and k-NN. The MARA has provided a good generalization performance of the RUS Boost prediction model.

Abstract In clinical practice, every decision should be reliable and explained to the stakeholders. The high accuracy of deep learning (DL) models pose a great advantage, but the fact that they function as black-boxes hinders their clinical applications. Hence, explainability methods became important as they provide explanation to DL models. In this study, two datasets with electrocardiogram (ECG) image representations of six heartbeats were built, one given the label of the last heartbeat and the other given the label of the first heartbeat. Each dataset was used to train one neural network. Finally, we applied well-known explainability methods to the resulting networks to explain their classifications. Explainability methods produced attribution maps where pixels intensities are proportional to their importance to the classification task. Then, we developed a metric to quantify the focus of the models in the heartbeat of interest. The classification models achieved testing accuracy scores of around 93.66% and 91.72%. The models focused around the heartbeat of interest,with values of the focus metric ranging between 8.8% and 32.4%. Future work will investigate the importance of regions outside the region of interest, besides the contribution of specific ECG waves to the classification.

Abstract This study aims to reflect on a list of libraries providing decision support to AI models. The goal is to assist in finding suitable libraries that support visual explainability and interpretability of the output of their AI model. Especially in sensitive application areas, such as medicine, this is crucial for understanding the decision-making process and for a safe application. Therefore, we use a glioma classification model’s reasoning as an underlying case. We present a comparison of 11 identified Python libraries that provide an addition to the better known SHAP and LIME libraries for visualizing explainability. The libraries are selected based on certain attributes, such as being implemented in Python, supporting visual analysis, thorough documentation, and active maintenance. We showcase and compare four libraries for global interpretations (ELI5, Dalex, InterpretML, and SHAP) and three libraries for local interpretations (Lime, Dalex, and InterpretML). As use case, we process a combination of openly available data sets on glioma for the task of studying feature importance when classifying the grade II, III, and IV brain tumor subtypes glioblastoma multiforme (GBM), anaplastic astrocytoma (AASTR), and oligodendroglioma (ODG), out of 1276 samples and 252 attributes. The exemplified model confirms known variations and studying local explainability contributes to revealing less known variations as putative biomarkers. The full comparison spreadsheet and implementation examples can be found in the appendix.

Abstract Deep learning techniques, such as convolutional neural networks (CNNs), generative adversarial networks (GANs), and graph neural networks (GNNs) have, over the past decade, changed the accuracy of prediction in many diverse fields. In recent years, the application of deep learning techniques in computer vision tasks in pathology has demonstrated extraordinary potential in assisting clinicians, automating diagnoses, and reducing costs for patients. Formerly unknown pathological evidence, such as morphological features related to specific biomarkers, copy number variations, and other molecular features, could also be captured by deep learning models. In this paper, we review popular deep learning methods and some recent publications about their applications in pathology.

Abstract The RNA polymerase III–specific TFIIIB complex is targeted by oncogenes and tumor suppressors, specifically the TFIIIB subunits BRF1, BRF2, and TBP. Currently, it is unclear if the TFIIIB subunit BDP1 is universally deregulated in human cancers. We performed a meta-analysis of patient data in the Oncomine database to analyze BDP1 alterations in human cancers. Herein, we report a possible role for BDP1 in non-Hodgkin’s lymphoma (NHL) for the first time. To the best of our knowledge, this is the first study to report a statistically significant decrease in BDP1 expression in patients with anaplastic lymphoma kinase–positive (ALK+) anaplastic large-cell lymphoma (ALCL) (p = 1.67 × 10−6 ) and Burkitt’s lymphoma (BL) (p = 1.54 × 10−11). Analysis of the BDP1 promoter identified putative binding sites for MYC, BCL6, E2F4, and KLF4 transcription factors, which were previously demonstrated to be deregulated in lymphomas. MYC and BDP1 expression were inversely correlated in ALK+ ALCL, suggesting a possible mechanism for the significant and specific decrease in BDP1 expression. In activated B-cell (ABC) diffuse large B-cell lymphoma (DLBCL), decreased BDP1 expression correlated with clinical outcomes, including recurrence at 1 year (p = 0.021) and 3 years (p = 0.005). Mortality at 1 (p = 0.030) and 3 (p = 0.012) years correlated with decreased BDP1 expression in ABC DLBCL. Together, these data suggest that BDP1 alterations may be of clinical significance in specific NHL subtypes and warrant further investigation.

Abstract Previous studies demonstrate the feasibility of predicting obesity using various machine learning techniques; however, these studies do not address the limitations of these methods in real-life settings where available data for children may vary. We investigated the medical history required for machine learning models to accurately predict body mass index (BMI) during early childhood. Within a longitudinal dataset of children ages 0–4 years, we developed predictive models based on long short-term memory (LSTM), a recurrent neural network architecture, using history EHR data from 2 to 8 clinical encounters to estimate child BMI. We developed separate, sex-stratified models using 80% of the data for training and 20% for external validation. We evaluated model performance using K-fold cross-validation, mean average error (MAE), and Pearson’s correlation coefficient (R2). Two history encounters and a 4-month prediction yielded a high prediction error and low correlation between predicted and actual BMI (MAE of 1.60 for girls and 1.49 for boys). Model performance improved with additional history encounters; improvement was not significant beyond five history encounters. The combined model outperformed the sex-stratified models, with a MAE = 0.98 (SD 0.03) and R2 = 0.72. Our models show that five history encounters are sufficient to predict BMI prior to age 4 for both boys and girls. Moreover, starting from an initial dataset with more than 269 exposure variables, we were able to identify a limited set of 24 variables that can facilitate BMI prediction in early childhood. Nine of these final variables are collected once, and the remaining 15 need to be updated during each visit.

Abstract Polysomnography is the gold-standard method for measuring sleep but is inconvenient and limited to a laboratory or a hospital setting. As a result, the vast majority of patients do not receive a proper diagnosis. In an attempt to solve this issue, sleep experts are continually looking for unobtrusive and affordable alternatives that can provide longitudinal sleep tracking. Collecting longitudinal data on sleep can accelerate epidemiological studies exploring the effect of sleep on health and disease. These alternatives can be in the form of wearables (e.g., actigraphs) or nonwearable (e.g., under-mattress sleep trackers). To this end, this paper aims to review the several attempts made by researchers toward unobtrusive sleep monitoring, specifically sleep cycle. We have performed a literature search between 2016 and 2021 and the following databases were used for retrieving related articles to unobtrusive sleep cycle monitoring: IEEE, Google Scholar, Journal of Clinical Sleep Medicine (JCSM), and PubMed Central (PMC). Following our survey, although existing devices showed promising results, most of the studies are restricted to a small sample of healthy individuals. Therefore, a broader scope of participants should be taken into consideration during future proposals and assessments of sleep cycle tracking systems. This is because factors such as gender, age, profession, and social class can largely affect sleep quality. Furthermore, a combination of sensors, e.g., smartwatches and under-mattress sleep trackers, are necessary to achieve reliable results. That is, wearables and nonwearable devices are complementary to each other, and so both are needed to boost the field of at-home sleep monitoring.

Abstract Metformin is the most used oral anti-diabetic drug in the world and consequently is commonly found in the aquatic environment. Some studies demonstrated that metformin may act as an endocrine-disrupting-chemical (EDC) in fish, although it does not have a classic EDC structure. In this sense, the aim of this work was to evaluate the potential disrupting effect of metformin in the cardiovascular system through in vitro, ex vivo, and in silico studies. For this purpose, human umbilical artery (HUA) and rat aorta artery (RAA) were used. The toxic concentrations of metformin were determined by a cytotoxicity assay and in silico simulations were performed to analyze the interactions of metformin with hormonal receptors. Our results show that metformin decreases viability of the smooth muscle cells. Moreover, metformin induces a vasorelaxant effect in rat aorta and human models by an endothelium-dependent and -independent pathways. Furthermore, docking simulations showed that metformin binds to androgen receptors (AR) and estrogen receptors (ERα and ERβ). In conclusion, the in silico assays suggested that metformin has the potential to be an endocrine disruptor, acting mainly on ERα. Further studies are needed to use metformin in pregnant women without impairing the cardiovascular health of the future generation.

Abstract This paper describes the study of overlapping leukaemia cells based on geometric features for identification and classification. Geometric features of blood cells are proposed to identify and classify overlapping cells into groups based on different overlapping degrees and the number of overlapped cells. In the proposed method, the percentage of average accuracy for identifying overlapping cells reached 98 percent. The proposed method can segment white blood cells from the overlapping cells with an overlapping degree of 70 percent. Improved Watershed Algorithm successfully increased 36.89 percent of accuracy in WBC segmentation. It reduced 46.12 percent of the over-segmentation problem. Tests of cell counting are conducted in the two stages, which are before and after the process of identification and classification of overlapping cells. The average percentage of total cell count is 83.31 percent, the average percentage of WBC counting is 84.8 percent, and the average percentage of RBC counting is 60.55 percent. The proposed method is efficient in identifying and classifying overlapping cells for increasing the accuracy of cell counting.

Abstract Motivation, BRCA1 and BRCA2 are genes with tumor suppressor activity. They are involved in a considerable number of biological processes. To help the biologist in tumor classification, we developed a deep learning algorithm. The question when we want to construct a neural network is how many hidden layers and neurons should we use. If the number of inputs and outputs is defined by the problem, the number of hidden layers and neurons is difficult to define. Hidden layers and neurons that make up each layer of the neural network influence the performance of system predictions. There are different methods for finding the optimal architecture. In this paper, we present the two packages that we have developed, the genetic algorithm (GA) and the particle swarm optimization (PSO) to optimize the parameters of the neural network for predicting BRCA1 and BRCA2 pathogenicity; Results, we will compare the results obtained by the two algorithms. We used datasets collected from our NGS analysis of BRCA1 and BRCA2 genes to train deep learning models. It represents a data collection of 11,875 BRCA1 and BRCA2 variants. Our preliminary results show that the PSO provided the most significant architecture of hidden layers and the number of neurons compared to grid search and GA; Conclusions, the optimal architecture found by the PSO algorithm is composed of 6 hidden layers with 275 hidden nodes with an accuracy of 0.98, precision 0.99, recall 0.98, and a specificity of 0.99.

Abstract Follicular lymphoma (FL) is one of the most frequent subtypes of non-Hodgkin lymphomas. This research predicted the prognosis of 184 untreated follicular lymphoma patients (LLMPP GSE16131 series), using gene expression data and artificial intelligence (AI) neural networks. A new strategy based on the random number generation was used to create 120 different and independent multilayer perceptron (MLP) solutions, and 22,215 gene probes were ranked according to their averaged normalized importance for predicting the overall survival. After dimensionality reduction, the final neural network architecture included (1) newly identified predictor genes related to cell adhesion and migration, cell signaling, and metabolism (EPB41L4B, MOCOS, SPIN2A, BTD, SRGAP3, CTNS, PRB1, L1CAM, and CEP57); (2) the international prognostic index (IPI); and (3) other relevant immuno-oncology, immune microenvironment, and checkpoint markers (CD163, CSF1R, FOXP3,PDCD1, TNFRSF14 (HVEM), and IL10). The performance of this neural network was good, with an area under the curve (AUC) of 0.89. A comparison with other machine learning techniques (C5 tree, logistic regression, Bayesian network, discriminant analysis, KNN algorithms, LSVM, random trees, SVM, tree-AS, XGBoost linear, XGBoost tree, CHAID, Quest, C&R tree, random forest, and neural network) was also made. In conclusion, the overall survival of follicular lymphoma was predicted with a neural network with high accuracy.