A Privacy-Preserving Approach to Effectively Utilize Distributed Data for Malaria Image Detection

Malaria is one of the life-threatening diseases caused by the parasite known as Plasmodium falciparum, affecting the human red blood cells. Therefore, it is an important to have an effective computer-aided system in place for early detection and treatment. The visual heterogeneity of the malaria dataset is highly complex and dynamic, therefore higher number of images are needed to train the machine learning (ML) models effectively. However, hospitals as well as medical institutions do not share the medical image data for collaboration due to general data protection regulations (GDPR) and the data protection act (DPA). To overcome this collaborative challenge, our research utilised real-time medical image data in the framework of federated learning (FL). We have used state-of-the-art ML models that include the ResNet-50 and DenseNet in a federated learning framework. We have experimented both models in different settings on a malaria dataset constituting 27,560 publicly available images and our preliminary results showed that the DenseNet model performed better in accuracy (75%) in contrast to ResNet-50 (72%) while considering eight clients, while the trend was observed as common in four clients with the similar accuracy of 94%, and six clients showed that the DenseNet model performed quite well with the accuracy of 92%, while ResNet-50 achieved only 72%. The federated learning framework enhances the accuracy due to its decentralised nature, continuous learning, and effective communication among clients, as well as the efficient local adaptation. The use of federated learning architecture among the distinct clients for ensuring the data privacy and following GDPR is the contribution of this research work.

The Time-Robustness Analysis of Individual Identification Based on Resting-State EEG

An ongoing interest towards identification based on biosignals, such as electroencephalogram (EEG), magnetic resonance imaging (MRI), is growing in the past decades. Previous studies indicated that the inherent information about brain activity may be used to identify individual during resting-state of eyes open (REO) and eyes closed (REC). Electroencephalographic (EEG) records the data from the scalp, and it is believed that the noisy EEG signals can influence the accuracies of one experiment causing unreliable results. Therefore, the stability and time-robustness of inter-individual features can be investigated for the purpose of individual identification. In this work, we conducted three experiments with the time interval of at least 2 weeks, and used different types of measures (Power Spectral Density, Cross Spectrum, Channel Coherence and Phase Lags) to extract the individual features. The Pearson Correlation Coefficient (PCC) is calculated to measure the level of linear correlation for intra-individual, and Support Vector Machine (SVM) is used to obtain the related classification accuracy. Results show that the classification accuracies of four features were 85-100% for intra-experiment dataset, and were 80-100% for fusion experiments dataset. For inter-experiments classification of REO features, the optimized frequency range is 13-40 Hz for three features, Power Spectral Density, Channel Coherence and Cross Spectrum. For inter-experiments classification of REC, the optimized frequency range is 8-40 Hz for three features, Power Spectral Density, Channel Coherence and Cross Spectrum. The classification results of Phase Lags are much lower than the other three features. These results show the time-robustness of EEG, which can further use for individual identification system.

Deep learning for early detection of pathological changes in X-ray bone microstructures: case of osteoarthritis

Texture features are designed to quantitatively evaluate patterns of spatial distribution of image pixels for purposes of image analysis and interpretation. Unexplained variations in the texture patterns often lead to misinterpretation and undesirable consequences in medical image analysis. In this paper we explore the ability of machine learning (ML) methods to design a radiology test of Osteoarthritis (OA) at early stage when the number of patients’ cases is small. In our experiments we use high-resolution X-ray images of knees in patients which were identified with Kellgren–Lawrence scores progressing from 1. The existing ML methods have provided a limited diagnostic accuracy, whilst the proposed Group Method of Data Handling strategy of Deep Learning has significantly extended the diagnostic test. The comparative experiments demonstrate that the proposed framework using the Zernike-based texture features has significantly improved the diagnostic accuracy on average by 11%. This allows us to conclude that the designed model for early diagnostic of OA will provide more accurate radiology tests, although new study is required when a large number of patients’ cases will be available.

A New dispersion entropy and fuzzy logic system methodology for automated classification of dementia stages using electroencephalograms

A new EEG-based methodology is presented for differential diagnosis of the Alzheimer's disease (AD), Mild Cognitive Impairment (MCI), and healthy subjects employing the discrete wavelet transform (DWT), dispersion entropy index (DEI), a recently-proposed nonlinear measurement, and a fuzzy logic-based classification algorithm. The effectiveness and usefulness of the proposed methodology are evaluated by employing a database of measured EEG data acquired from 135 subjects, 45 MCI, 45 AD and 45 healthy subjects. The proposed methodology differentiates MCI and AD patients from HC subjects with an accuracy of 82.6-86.9%, sensitivity of 91 %, and specificity of 87 %.

Person identification from EEG using various machine learning techniques with inter-hemispheric amplitude ratio

Association between electroencephalography (EEG) and individually personal information is being explored by the scientific community. Though person identification using EEG is an attraction among researchers, the complexity of sensing limits using such technologies in real-world applications. In this research, the challenge has been addressed by reducing the complexity of the brain signal acquisition and analysis processes. This was achieved by reducing the number of electrodes, simplifying the critical task without compromising accuracy. Event-related potentials (ERP), a.k.a. time-locked stimulation, was used to collect data from each subject’s head. Following a relaxation period, each subject was visually presented a random four-digit number and then asked to think of it for 10 seconds. Fifteen trials were conducted with each subject with relaxation and visual stimulation phases preceding each mental recall segment. We introduce a novel derived feature, dubbed Inter-Hemispheric Amplitude Ratio (IHAR), which expresses the ratio of amplitudes of laterally corresponding electrode pairs. The feature was extracted after expanding the training set using signal augmentation techniques and tested with several machine learning (ML) algorithms, including Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and k-Nearest Neighbor (kNN). Most of the ML algorithms showed 100% accuracy with 14 electrodes, and according to our results, perfect accuracy can also be achieved using fewer electrodes. However, AF3, AF4, F7, and F8 electrode combination with kNN classifier which yielded 99.0±0.8% testing accuracy is the best for person identification to maintain both user-friendliness and performance. Surprisingly, the relaxation phase manifested the highest accuracy of the three phases.

Multiparametric quantitative and texture 18F-FDG PET/CT analysis for primary malignant tumour grade differentiation

Background

¹⁸F-FDG positron emission tomography/computed tomography (PET/CT) is a successfully used imaging modality in oncology. The aim of the study was to investigate a connection of epithelial tumour differentiation grade with both semiquantitative and quantitative metabolic PET data focusing on creation of multiparametric model of tumour grade prediction utilising both standardised uptake value-based and texture-based ¹⁸F-FDG PET parameters and to investigate an influence of different image segmentation techniques on these parameters and modelling.

Methods

¹⁸F-FDG PET/CT data from 84 patients with epithelial malignant tumours was retrospectively analysed to create sets of both conventional semiquantitative (based on standardised uptake values), volumetric, and quantitative texture metabolic parameters of primary tumours with four different segmentation techniques.

Results

Most of the calculated volumetric and texture parameters showed to be influenced by segmentation technique. There was no significant difference in values of only three parameters, in all four segmentation methods: homogeneity, energy, and sphericity. Almost every extracted parameter in all segmentation technique subsets showed significant ability to discriminate individual tumour grade versus the subset of remaining two tumour grades. No parameters were able to discriminate all three tumour grades separately simultaneously or without the overlapping of threshold values. Group method of data handling (GMDH) modelling included all the above-mentioned extracted parameters. The highest value to discriminate tumour grade was achieved using ITK-SNAP segmentation, with an accuracy ranging from 91 to 100%.

Conclusions

Multiparametric modelling with GMDH utilising both semiquantitative and quantitative texture metabolic PET parameters seems to be an interesting tool for non-invasive malignant epithelial tumours grade differentiation.

Decrypting the electrophysiological individuality of the human brain: Identification of individuals based on resting-state EEG activity

Biometric identification (BI) of individuals is a fast-growing field of research that is producing increasingly sophisticated applications in several spheres of everyday life. Previous magnetic resonance imaging (MRI) studies have demonstrated that based on the high inter-individual variability of brain structure and function, it is possible to identify individuals with high accuracy. Otherwise, there is the common belief that electroencephalographic (EEG) data recorded at the surface of the scalp are too noisy for identification purposes with a comparably high hit rate. In the present work, we compared BI quality (F1-scores, accuracy, sensitivity, and specificity) between different types of functional (instantaneous, lagged, and total coherence, phase synchronization, correlation, and mutual information) and effective (Granger causality, phase synchronization, and coherence) connectivity measures. Results revealed that across functional connectivity metrics, identification accuracy was in the range of 0.98-1, whereas sensitivity and F1-scores were between 0.00 and 1 and specificity was between 0.99 and 1. BI was higher for the connectivity metrics that are contaminated by volume conduction (instantaneous connectivity) compared to those that are unaffected by this variable (lagged connectivity). Support vector machine and neural network algorithms yielded the highest BI, followed by random forest and weighted k-nearest neighborhood, whereas linear discriminant analysis was less accurate. These results provide cross-validated counterevidence to the belief that EEG data are too noisy for identification purposes and demonstrate that functional and effective connectivity metrics are particularly suited for BI applications with comparable accuracy to MRI. Our results have important implications for fast, low-cost, and mobile BI applications.