Fraunhofer Portugal Research

Barandas M., Folgado D., Fernandes L., Santos S., Abreu M., Bota P., Liu H., Schultz T., Gamboa H.

Time series feature extraction is one of the preliminary steps of conventional machine learning pipelines. Quite often, this process ends being a time consuming and complex task as data scientists must consider a combination between a multitude of domain knowledge factors and coding implementation. We present in this paper a Python package entitled Time Series Feature Extraction Library (TSFEL), which computes over 60 different features extracted across temporal, statistical and spectral domains. User customisation is achieved using either an online interface or a conventional Python package for more flexibility and integration into real deployment scenarios. TSFEL is designed to support the process of fast exploratory data analysis and feature extraction on time series with computational cost evaluation.

Martins J., Cardoso J.S., Soares F.

Background and Objective: Glaucoma, an eye condition that leads to permanent blindness, is typically asymptomatic and therefore difficult to be diagnosed in time. However, if diagnosed in time, Glaucoma can effectively be slowed down by using adequate treatment; hence, an early diagnosis is of utmost importance. Nonetheless, the conventional approaches to diagnose Glaucoma adopt expensive and bulky equipment that requires qualified experts, making it difficult, costly and time-consuming to diagnose large amounts of people. Consequently, new alternatives to diagnose Glaucoma that suppress these issues should be explored. Methods: This work proposes an interpretable computer-aided diagnosis (CAD) pipeline that is capable of diagnosing Glaucoma using fundus images and run offline in mobile devices. Several public datasets of fundus images were merged and used to build Convolutional Neural Networks (CNNs) that perform segmentation and classification tasks. These networks are then used to build a pipeline for Glaucoma assessment that outputs a Glaucoma confidence level and also provides several morphological features and segmentations of relevant structures, resulting in an interpretable Glaucoma diagnosis. To assess the performance of this method in a restricted environment, this pipeline was integrated into a mobile application and time and space complexities were assessed. Results: Considering the test set, the developed pipeline achieved 0.91 and 0.75 of Intersection over Union (IoU) in the optic disc and optic cup segmentation, respectively. With regards to the classification, an accuracy of 0.87 with a sensitivity of 0.85 and an AUC of 0.93 were attained. Moreover, this pipeline runs on an average Android smartphone in under two seconds. Conclusions: The results demonstrate the potential that this method can have in the contribution to an early Glaucoma diagnosis. The proposed approach achieved similar or slightly better metrics than the current CAD systems for Glaucoma assessment while running on more restricted devices. This pipeline can, therefore, be used to construct accurate and affordable CAD systems that could enable large Glaucoma screenings, contributing to an earlier diagnose of this condition.

Adcock M., Fankhauser M., Post J., Lutz K., Zizlsperger L., Luft A.R., Guimarães V., Schättin A., de Bruin E.D.

Aging is associated with a decline in physical functions, cognition and brain structure. Considering that human life is based on an inseparable physical-cognitive interplay, combined physical-cognitive training through exergames is a promising approach to counteract age-related impairments. The aim of this study was to assess the effects of an in-home multicomponent exergame training on [i] physical and cognitive functions and [ii] brain volume of older adults compared to a usual care control group. Thirty-seven healthy and independently living older adults aged 65 years and older were randomly assigned to an intervention (exergame training) or a control (usual care) group. Over 16 weeks, the participants of the intervention group absolved three home-based exergame sessions per week (à 30-40 minutes) including Tai Chi-inspired exercises, dancing and step-based cognitive games. The control participants continued with their normal daily living. Pre- and post-measurements included assessments of physical (gait parameters, functional muscle strength, balance, aerobic endurance) and cognitive (processing speed, short-term attention span, working memory, inhibition, mental flexibility) functions. T1-weighted magnetic resonance imaging was conducted to assess brain volume. Thirty-one participants (mean age = 73.9 ± 6.4 years, range = 65 – 90 years, 16 female) completed the study. Inhibition and working memory significantly improved post-intervention in favour of the intervention group (inhibition: F(1) = 2.537, p = .046, n2p = .11, working memory: F(1) = 5.872, p = .015, n2p = .02). Two measures of short-term attentional span showed improvements after training in favor of the control group (F(1) = 4.309, p = .038, n2p = .03, F(1) = 8.504, p = .004, n2p = .04). No significant interaction effects were evident for physical functions or brain volume. Both groups exhibited a significant decrease in gray matter volume of frontal areas and the hippocampus over time. The findings indicate a positive influence of exergame training on executive functioning. No improvements in physical functions or brain volume were evident in this study. Better adapted individualized training challenge and a longer training period are suggested. Further studies are needed that assess training-related structural brain plasticity and its effect on performance, daily life functioning and healthy aging.

Mariano J., Marques S., Ramos M.R., Gerardo F., Cunha C.L., Girenko A., Alexandersson J., Stree B., Lamanna M., Lorenzatto M., Mikkelsen L.P., Bundgård-Jørgensen U., Rêgo S., de Vries H.

Older adults are often stereotyped as having less technological ability than younger age groups. As a result, older individuals may avoid using technology due to stereotype threat, the fear of conf...

Neves I., Folgado D., Santos S., Barandas M., Campagner A., Ronzio L., Cabitza F., Gamboa H.

Treatment and prevention of cardiovascular diseases often rely on Electrocardiogram (ECG) interpretation. Dependent on the physician's variability, ECG interpretation is subjective and prone to errors. Machine learning models are often developed and used to support doctors; however, their lack of interpretability stands as one of the main drawbacks of their widespread operation. This paper focuses on an Explainable Artificial Intelligence (XAI) solution to make heartbeat classification more explainable using several state-of-the-art model-agnostic methods. We introduce a high-level conceptual framework for explainable time series and propose an original method that adds temporal dependency between time samples using the time series' derivative. The results were validated in the MIT-BIH arrhythmia dataset: we performed a performance's analysis to evaluate whether the explanations fit the model's behaviour; and employed the 1-D Jaccard's index to compare the subsequences extracted from an interpretable model and the XAI methods used. Our results show that the use of the raw signal and its derivative includes temporal dependency between samples to promote classification explanation. A small but informative user study concludes this study to evaluate the potential of the visual explanations produced by our original method for being adopted in real-world clinical settings, either as diagnostic aids or training resource. • We present an in-depth study on the technical feasibility and practical usefulness of visual explanations for ECG classifiers • We propose using the time series derivate to support state-of-the-art XAI methods measuring feature importance considering the temporal domain • We conducted an informative user study to evaluate the potential of visual explanations on ECGs

Santos M.S., Abreu P.H., Japkowicz N., Fernández A., Soares C., Wilk S., Santos J.

Current research on imbalanced data recognises that class imbalance is aggravated by other data intrinsic characteristics, among which class overlap stands out as one of the most harmful. The combination of these two problems creates a new and difficult scenario for classification tasks and has been discussed in several research works over the past two decades. In this paper, we argue that despite some insightful information can be derived from related research, the joint-effect of class overlap and imbalance is still not fully understood, and advocate for the need to move towards a unified view of the class overlap problem in imbalanced domains. To that end, we start by performing a thorough analysis of existing literature on the joint-effect of class imbalance and overlap, elaborating on important details left undiscussed on the original papers, namely the impact of data domains with different characteristics and the behaviour of classifiers with distinct learning biases. This leads to the hypothesis that class overlap comprises multiple representations, which are important to accurately measure and analyse in order to provide a full characterisation of the problem. Accordingly, we devise two novel taxonomies, one for class overlap measures and the other for class overlap-based approaches, both resonating with the distinct representations of class overlap identified. This paper therefore presents a global and unique view on the joint-effect of class imbalance and overlap, from precursor work to recent developments in the field. It meticulously discusses some concepts taken as implicit in previous research, explores new perspectives in light of the limitations found, and presents new ideas that will hopefully inspire researchers to move towards a unified view on the problem and the development of suitable strategies for imbalanced and overlapped domains.

Lopes P., Silva E., Braga C., Oliveira T., Rosado L.

The lack of transparency of powerful Machine Learning systems paired with their growth in popularity over the last decade led to the emergence of the eXplainable Artificial Intelligence (XAI) field. Instead of focusing solely on obtaining highly performing models, researchers also develop explanation techniques that help better understand the system’s reasoning for a particular output. An explainable system can be designed, developed, and evaluated from different perspectives, which enables researchers from different disciplines to work together on this topic. However, the multidisciplinary nature of XAI systems creates new challenges for condensing and structuring adequate methodologies to design and evaluate such systems. This paper presents a survey of Human-centred and Computer-centred methods to evaluate XAI systems. We propose a new taxonomy to categorize XAI evaluation methods more clearly and intuitively. This categorization gathers knowledge from different disciplines and organizes the evaluation methods according to a set of categories that represent key properties of XAI systems. Possible ways to use the proposed taxonomy in the design and evaluation of XAI systems are also discussed, alongside with some concluding remarks and future directions of research.

Cabitza F., Campagner A., Ronzio L., Cameli M., Mandoli G.E., Pastore M.C., Sconfienza L.M., Folgado D., Barandas M., Gamboa H.

In this paper, we study human-AI collaboration protocols, a design-oriented construct aimed at establishing and evaluating how humans and AI can collaborate in cognitive tasks. We applied this construct in two user studies involving 12 specialist radiologists (the knee MRI study) and 44 ECG readers of varying expertise (the ECG study), who evaluated 240 and 20 cases, respectively, in different collaboration configurations. We confirm the utility of AI support but find that XAI can be associated with a “white-box paradox”, producing a null or detrimental effect. We also find that the order of presentation matters: AI-first protocols are associated with higher diagnostic accuracy than human-first protocols, and with higher accuracy than both humans and AI alone. Our findings identify the best conditions for AI to augment human diagnostic skills, rather than trigger dysfunctional responses and cognitive biases that can undermine decision effectiveness.

Resende C., Folgado D., Oliveira J., Franco B., Moreira W., Oliveira-Jr A., Cavaleiro A., Carvalho R.

Industry 4.0, allied with the growth and democratization of Artificial Intelligence (AI) and the advent of IoT, is paving the way for the complete digitization and automation of industrial processes. Maintenance is one of these processes, where the introduction of a predictive approach, as opposed to the traditional techniques, is expected to considerably improve the industry maintenance strategies with gains such as reduced downtime, improved equipment effectiveness, lower maintenance costs, increased return on assets, risk mitigation, and, ultimately, profitable growth. With predictive maintenance, dedicated sensors monitor the critical points of assets. The sensor data then feed into machine learning algorithms that can infer the asset health status and inform operators and decision-makers. With this in mind, in this paper, we present TIP4.0, a platform for predictive maintenance based on a modular software solution for edge computing gateways. TIP4.0 is built around Yocto, which makes it readily available and compliant with Commercial Off-the-Shelf (COTS) or proprietary hardware. TIP4.0 was conceived with an industry mindset with communication interfaces that allow it to serve sensor networks in the shop floor and modular software architecture that allows it to be easily adjusted to new deployment scenarios. To showcase its potential, the TIP4.0 platform was validated over COTS hardware, and we considered a public data-set for the simulation of predictive maintenance scenarios. We used a Convolution Neural Network (CNN) architecture, which provided competitive performance over the state-of-the-art approaches, while being approximately four-times and two-times faster than the uncompressed model inference on the Central Processing Unit (CPU) and Graphical Processing Unit, respectively. These results highlight the capabilities of distributed large-scale edge computing over industrial scenarios.

Rodrigues J., Liu H., Folgado D., Belo D., Schultz T., Gamboa H.

Biosignal-based technology has been increasingly available in our daily life, being a critical information source. Wearable biosensors have been widely applied in, among others, biometrics, sports, health care, rehabilitation assistance, and edutainment. Continuous data collection from biodevices provides a valuable volume of information, which needs to be curated and prepared before serving machine learning applications. One of the universal preparation steps is data segmentation and labelling/annotation. This work proposes a practical and manageable way to automatically segment and label single-channel or multimodal biosignal data using a self-similarity matrix (SSM) computed with signals’ feature-based representation. Applied to public biosignal datasets and a benchmark for change point detection, the proposed approach delivered lucid visual support in interpreting the biosignals with the SSM while performing accurate automatic segmentation of biosignals with the help of the novelty function and associating the segments grounded on their similarity measures with the similarity profiles. The proposed method performed superior to other algorithms in most cases of a series of automatic biosignal segmentation tasks; of equal appeal is that it provides an intuitive visualization for information retrieval of multimodal biosignals.