Dynamic Causal Modeling (DCM) is a Bayesian framework to investigate effective connectivity between brain regions using neuroimaging data. High noise levels can significantly affect the efficiency and reliability of DCM. Here, we propose a new multivariate method, called Multivariate Variational Mode Decomposition (MVMD) for enhanced DCM (MVMD-DCM). This method works on the extracted time-series of the regions of interest by reducing the contribution of noise, which ultimately ensures that only relevant task-related information in the time-series are fed to DCM. We demonstrate the effectiveness of MVMD-DCM in mitigating the impact of noise using simulated task-based fMRI data. Simulated data was generated at two Signal-to-Noise Ratio (SNR) levels of 0.5 (-3dB) and 1 (0dB). A comparative analysis with respect to model evidence, true model selection frequency and model parameters estimation demonstrated the superiority of MVMD-DCM over the default DCM for both SNR levels. Our study paves the way for the development of robust methods for inferring effective connectivity with DCM from noisy fMRI data.

Recently, there has been a growing interest in analyzing resting-state fMRI (rs-fMRI) signals using Time-Varying Phase Synchronization (TVPS) measures. TVPS serves as a functional connectivity metric, allowing for the quantification of phase synchronization between different brain regions. However, extracting the phase from fMRI signals poses challenges due to inherent noise and insufficient bandlimitation. Traditional filtering methods struggle to effectively eliminate noise and necessitate prior knowledge of cutoff frequencies. In this context, data-driven multivariate decomposition techniques present promising solutions for extracting narrow-band components suitable for TVPS analyses. Previous studies have identified Multivariate Variational Mode Decomposition (MVMD) as particularly suitable for fMRI decomposition. However, MVMD's requirement for predefining the number of extracted modes K limits its analytical capabilities. To address this limitation, we employ an enhanced MVMD scheme called Noise-Assisted MVMD (NA-MVMD), designed to reduce sensitivity to parameters and enhance decomposition quality. We apply NA-MVMD to synthetic signals, showcasing improved decomposition quality, noise robustness, and reduced sensitivity in setting the K parameter.

Despite the emergence of numerous Deep Learning (DL) models for breast cancer detection via mammograms, there is a lack of evidence about their robustness to perform well on new unseen mammograms. To fill this gap, we introduce StethoNet, a DL-based framework that consists of multiple Convolutional Neural Network (CNN) trained models for classifying benign and malignant tumors. StethoNet was trained on the Chinese Mammography Database (CMMD), and tested on unseen images from CMMD, as well as on images from two independent datasets, i.e., the Vindr-Mammo and the INbreast datasets. To mitigate domain-shift effects, we applied an effective entropy-based domain adaptation technique at the preprocessing stage. Furthermore, a Bayesian hyperparameters optimization scheme was implemented for StethoNet optimization. To ensure interpretable results that corroborate with prior clinical knowledge, attention maps generated using Gradient-weighted Class Activation Mapping (GRADCAM) were compared with Regions of Interest (ROIs) identified by radiologists. StethoNet achieved impressive Area Under the receiver operating characteristics Curve (AUC) scores: 90.7% (88.6%-92.8%), 83.9% (76.0%-91.8%), and 85.7% (82.1%-89.4%) for the CMMD, INbreast, and Vindr-Mammo datasets, respectively. These results surpass the current state of the art and highlight the robustness and generalizability of StethoNet, scaffolding the integration of DL models into breast cancer mammography screening workflows.

A document by Charalampos Lamprou . Click on the document to view its contents.

Objective: Rest tremor is one of the most common symptoms of Parkinson’s Disease (PD), with diagnosis and severity estimation often being hindered by the subjectivity of clinical methods and limitations of existing screening procedures. Hence, development of methods that can accurately describe properties of PD rest tremor, while accounting for the presence of possible ongoing treatments, such as Deep Brain Stimulation (DBS) and medication, are quite important. Methods: A Higher Order Spectrum (HOS)-based analysis for characterizing and classifying tremor severity and treatment effectiveness, taking into consideration its nonlinear characteristics, is proposed here. Bispectrum- and Bicoherence-based features are extracted from velocity signals recorded from 16 PD patients at their index finger. Two different scenarios are implemented and tested for feature characterization under various conditions of treatment. Moreover, a classification scheme was constructed to evaluate the ability of HOS-based features in accurately predicting the rest tremor level, i.e., Low Amplitude/High Amplitude (LAT/HAT), and the presence of treatment (medication/DBS On-Off). To avoid over-fitting, a leave-one-subject-out cross-validation procedure was adopted. Results: The proposed bispectral analysis resulted in area under the Receiver Operating Characteristics curve (AUC) score of 0.96 (95% Confidence Interval (CI): 0.89-1) with 0.91/0.88 Sensitivity/Specificity, respectively, for the On-Off classification for Medication treatment. For the DBS treatment, the proposed analysis resulted in an AUC score of 0.72 (95% CI: 0.57-0.87), with 0.64/0.65 Sensitivity/Specificity, respectively. For LAT/HAT prediction, the best performing models yield scores of 1/0.93 for AUC/Accuracy metrics, with 1/0.86 Sensitivity/Specificity, respectively. Conclusions: When compared to the existing methods, the proposed methodology outperforms them regarding the prediction of LAT/HAT and treatment effectiveness. Additionally, our HOS-based methodology enables the establishment of new rest tremor classes, based on its nonlinearity and allows for new insights about the dynamic nature of the resting tremor production system. Significance: We propose a method that can effectively recognize rest tremor severity and assess the influence of medication and DBS. This study introduces, for the first time, nonlinearity-based classes for rest tremor and provides an accurate representation of tremor dynamics through HOS.