1,720,992 research outputs found
Modelli di Intelligenza Artificiale per l'analisi di dati da neuroimaging multimodale
No full textMedical imaging (MI) refers to several technologies that provide images of organs and tissues of human body for diagnosis and scientific purposes. Furthermore, the technologies that allow us to capture medical images and signals are advancing rapidly, providing higher quality images of previously unmeasured biological features at decreasing costs. This has mainly occurred for highly specialized applications, such as cardiology and neurology.
Artificial Intelligence (AI), which to date has largely focused on non medical applications, such as computer vision, provides to be an instrumental toolkit that will help unleash the potential of MI. In fact, the significant variability in anatomy across individuals, the lack of specificity of the imaging techniques, the unpredictability of the diseases, the weakness of the biological signals, the presence of noise and artifacts and the complexities of the underlying biology often make it impossible to derive deterministic algorithmic solutions for the problems encountered in neurology.
Aim of this thesis was to develop AI models capable of carrying out quantitative, objective, accurate and reliable analyzes of imaging tools, EEG and MRI, used in neurology. Beyond the development of AI models, attention was focused on the quality of data which can be lowered by the "uncertainty" produced by the issues cited above. Further, the uncertainty affecting data was also described, discussed and addressed.
Main results have been the proposal of innovative AI-based strategies for signal and image improvement through artifact reduction and data stabilization both in EEG and in MRI. This has allowed to apply EEG for weak signals recognition and interpretation (infant 3M patients), to provide effective strategies for dealing MRI variability and uncertainty in multiple sclerosis segmentation, both for single source and multiple-source MRI. According to the used evaluation criteria, the obtained results are comparable with those obtained by human experts.
Future developments will regard the generalization of the proposed strategies to cope with different diseases or with different applications of MI. Particular attention will be paid to the optimization of the models and to understand the processes underlying their behavior. To this aim, specific strategies for checking the deep structures of the proposed architectures will be studied. In this way, besides model optimization, it would be possible to get the functional relationships among the features generating from the model and use them to improve human knowledge (a sort of inverse transfer learning)
A fast and scalable framework for automated artifact recognition from EEG signals represented in scalp topographies of Independent Components
No full textBACKGROUND AND OBJECTIVES: Electroencephalography (EEG) measures the electrical brain activity in real-time by using sensors placed on the scalp. Artifacts due to eye movements and blinking, muscular/cardiac activity and generic electrical disturbances, have to be recognized and eliminated to allow a correct interpretation of the Useful Brain Signals (UBS). Independent Component Analysis (ICA) is effective to split the signal into Independent Components (IC) whose re-projection on 2D topographies of the scalp (images also called Topoplots) allows to recognize/separate artifacts and UBS. Topoplot analysis, a gold standard for EEG, is usually carried out offline either visually by human experts or through automated strategies, both unenforceable when a fast response is required as in online Brain-Computer Interfaces (BCI). We present a fully automatic, effective, fast, scalable framework for artifacts recognition from EEG signals represented in IC Topoplots to be used in online BCI.METHODS: The proposed architecture, optimized to contain three 2D Convolutional Neural Networks (CNN), divides Topoplots in 4 classes: 3 types of artifacts and UBS. The framework architecture is described and the results are presented, discussed and indirectly compared with those obtained from state-of-the-art competitive strategies.RESULTS: Experiments on public EEG datasets showed overall accuracy, sensitivity and specificity greater than 98%, taking 1.4s on a standard PC for 32 Topoplots, i.e. for an EEG system with at least 32 sensors.CONCLUSIONS: The proposed framework is faster than other automatic methods based on IC analysis and fast enough to be used in EEG-based online BCI. In addition, its scalable architecture and ease of training are necessary conditions to apply it in BCI, where difficult operating conditions caused by uncontrolled muscle spasms, eye rotations or head movements, produce specific artifacts that need to be recognized and dealt with
CNN-based artifact recognition from independent components of EEG signals
No full textElectroencephalography measures brain activity in real time. Artifacts are spurious signals due to eye movements and blinking, muscular/cardiac activity, and generic electrical disturbances. Artifacts have to be recognized and eliminated to allow a correct interpretation of the useful brain signals (UBS). Independent component analysis (ICA) is an effective strategy to retrieve the independent components (IC) of the signals. The reprojection of those components on 2D topographies of the scalp (topoplots) facilitates the recognition of artifacts from UBS. In this chapter, we describe the preprocessing pipeline based on the topoplot analysis of ICs. Several ICA algorithms are introduced, both for offline and online artifact recognition. In particular, we discuss several automatic methods that, emulating the human vision approach, are capable of recognizing all the common artifacts with the same precision as human experts. Finally, as a case study, we chose one of them and described its pipeline in detail. The chosen architecture, composed of a modular ensemble of 2D convolutional neural networks, is capable of recognizing all the most common artifacts in an online mode
Autonomous Driving: Integration of Segmentation and Depth Camera in a Curriculum Learning Approach
No full textSiamese Network to Investigate Scanner-Dependency in MRI
No full textMagnetic resonance imaging (MRI) is an effective imaging tool that, due to its non-invasiveness and multiple-parameter nature, is frequently used in medicine. In particular, the MRI's inherent flexibility deriving from the usage of multiple parameters allows to obtain images of variable contrast and quality. However, intrinsic MRI contrast variability often comes with drawbacks in terms of differences in different scanners, thus resulting in the impossibility of standardizing the image contrast. In particular, this variability could negatively affect the automatic analysis of Deep Learning (DL) methods, both in the training phase and in the test phase. In this work, we present several results on how images collected from different MRI scanners are handled by DL methods. To this end, we trained a Siamese network (SNN), based on the EfficientNet-B0 Convolutional Neural Network (EN-CNN), to learn how to recognize the scanner that has generated a given image. The output encoding features of the SNN have been projected into a 2D space with Uniform Manifold Approximation and Projection (UMAP) and have been discussed. Regarding the training phase, the UMAP projects show that the network is capable of separating MR images encoded features from different MRI scanners. Moreover, even if the MR images of different subjects are acquired with the same scanner, the results suggest that there are considerable differences in how the SNN encoded those features. The test phase confirmed that the SNN architecture is capable of recognizing images from different MRI scanners
A light CNN for detecting COVID-19 from CT scans of the chest
No full textComputer Tomography (CT) imaging of the chest is a valid diagnosis tool to detect COVID-19 promptly and to control the spread of the disease. In this work we propose a light Convolutional Neural Network (CNN) design, based on the model of the SqueezeNet, for the efficient discrimination of COVID-19 CT images with respect to other community-acquired pneumonia and/or healthy CT images. The architecture allows to an accuracy of 85.03% with an improvement of about 3.2% in the first dataset arrangement and of about 2.1% in the second dataset arrangement. The obtained gain, though of low entity, can be really important in medical diagnosis and, in particular, for Covid-19 scenario. Also the average classification time on a high-end workstation, 1.25 s, is very competitive with respect to that of more complex CNN designs, 13.41 s, witch require pre-processing. The proposed CNN can be executed on medium-end laptop without GPU acceleration in 7.81 s: this is impossible for methods requiring GPU acceleration. The performance of the method can be further improved with efficient pre-processing strategies for witch GPU acceleration is not necessary. (C) 2020 Elsevier B.V. All rights reserved
Measurements by A LEAP-Based Virtual Glove for the hand rehabilitation
Full text linkHand rehabilitation is fundamental after stroke or surgery. Traditional rehabilitation
requires a therapist and implies high costs, stress for the patient, and subjective evaluation of
the therapy effectiveness. Alternative approaches, based on mechanical and tracking-based gloves,
can be really effective when used in virtual reality (VR) environments. Mechanical devices are often
expensive, cumbersome, patient specific and hand specific, while tracking-based devices are not
affected by these limitations but, especially if based on a single tracking sensor, could suffer from
occlusions. In this paper, the implementation of a multi-sensors approach, the Virtual Glove (VG),
based on the simultaneous use of two orthogonal LEAP motion controllers, is described. The VG is
calibrated and static positioning measurements are compared with those collected with an accurate
spatial positioning system. The positioning error is lower than 6 mm in a cylindrical region of interest
of radius 10 cm and height 21 cm. Real-time hand tracking measurements are also performed, analysed
and reported. Hand tracking measurements show that VG operated in real-time (60 fps), reduced
occlusions, and managed two LEAP sensors correctly, without any temporal and spatial discontinuity
when skipping from one sensor to the other. A video demonstrating the good performance of VG
is also collected and presented in the Supplementary Materials. Results are promising but further
work must be done to allow the calculation of the forces exerted by each finger when constrained by
mechanical tools (e.g., peg-boards) and for reducing occlusions when grasping these tools. Although
the VG is proposed for rehabilitation purposes, it could also be used for tele-operation of tools and
robots, and for other VR applications
Forces calculation module for the leap-based virtual glove
No full textHand rehabilitation is fundamental after stroke or surgery. Traditional rehabilitation implies high costs, stress for the patient, and subjective evaluation of the therapy effectiveness. Mechanical devices based approaches are often expensive, cumbersome and patient specific, while tracking-based devices are not affected by these limitations, though they could suffer from occlusions. In recent works, the procedure used for implementing a multi-sensors approach, the Virtual Glove (VG), based on the simultaneous use of two orthogonal LEAP motion controllers, was described. In this paper, an engineered version of VG was calibrated and measurements were performed. This article presents a model extension to be used for the off-line calculation of the hand kinematics and of the flexion/extension forces exerted by each finger when constrained by calibrated elastic tools
Unsupervised Brain MRI Anomaly Detection for Multiple Sclerosis Classification
No full textSupervised deep learning has been widely applied in medical imaging to detect multiple sclerosis. However, it is difficult to have perfectly annotated lesions in magnetic resonance images, due to the inherent difficulties with the annotation process performed by human experts. To provide a model that can completely ignore annotations, we propose an unsupervised anomaly detection approach. The method uses a convolutional autoencoder to learn a "normal brain" distribution and detects abnormalities as a deviation from the norm. Experiments conducted with the recently released OASIS-3 dataset and the challenging MSSEG dataset show the feasibility of the proposed method, as very encouraging sensitivity and specificity were achieved in the binary health/disease discrimination. Following the "normal brain" learning rule, the proposed approach can easily generalize to other types of brain diseases, due to its potential to detect arbitrary anomalies
Siamese network to assess scanner-related contrast variability in MRI
Full text linkMagnetic Resonance Imaging (MRI) stands as a noninvasive tool for diagnosing and monitoring various diseases. The flexibility of MRI configuration parameters allows for adaptable imaging sequences, and at the same time poses challenges in terms of reproducibility, as variability in imaging sequences leads to significant differences in image contrast. This is one of the major causes that compromise the reliability of deep learning methods. Since the majority of the literature is focused on documenting the effects of this issue rather than delving into its underlying causes, this work follows a different approach. A Siamese Neural Network (SNN) has been trained to identify the scanner that acquired the input image. Experimental results include the use of Euclidean Distance (ED) and machine learning algorithms trained and tested using the feature vectors generated with the SNN. The results have shown that the proposed method is capable of distinguishing the scanner used for the acquisition with high accuracy. For a comprehensive interpretation of the results, the feature vectors have been dimensionality reduced and visualized with a 3D plot. Finally, the proposed method is sensitive to MR image contrast variability and could be used to detect data-related inconsistencies and provide a mechanism to make users aware of potential issues
- …
