Increasingly complex learning methods such as boosting, bagging and deep learning have made ML models more accurate, but harder to interpret and explain, culminating in black-box machine learning models. Model developers and users alike are often presented with a trade-off between performance and intelligibility, especially in high-stakes applications like medicine. In the present article we propose a novel methodological approach for generating explanations for the predictions of a generic machine learning model, given a specific instance for which the prediction has been made. The method, named AraucanaXAI, is based on surrogate, locally-fitted classification and regression trees that are used to provide post-hoc explanations of the prediction of a generic machine learning model. Advantages of the proposed XAI approach include superior fidelity to the original model, ability to deal with non-linear decision boundaries, and native support to both classification and regression problems. We provide a packaged, open-source implementation of the AraucanaXAI method and evaluate its behaviour in a number of different settings that are commonly encountered in medical applications of AI. These include potential disagreement between the model prediction and physician's expert opinion and low reliability of the prediction due to data scarcity.

Designing effective theory-driven digital behaviour change interventions (DBCI) is a challenging task. To ease the design process, and assist with knowledge sharing and evaluation of the DBCI, we propose the SATO (IDEAS expAnded wiTh BCIO) design workflow based on the IDEAS (Integrate, Design, Assess, and Share) framework and aligned with the Behaviour Change Intervention Ontology (BCIO). BCIO is a structural representation of the knowledge in behaviour change domain supporting evaluation of behaviour change interventions (BCIs) but it is not straightforward to utilise it during DBCI design. IDEAS (Integrate, Design, Assess, and Share) framework guides multi-disciplinary teams through the mobile health (mHealth) application development life-cycle but it is not aligned with BCIO entities. SATO couples BCIO entities with workflow steps and extends IDEAS Integrate stage with consideration of customisation and personalisation. We provide a checklist of the activities that should be performed during intervention planning with concrete examples and a tutorial accompanied with case studies from the Cancer Better Life Experience (CAPABLE) European project. In the process of creating this workflow, we found the necessity to extend the BCIO to support the scenarios of multiple clinical goals in the same application. To ensure the SATO steps are easy to follow for the incomers to the field, we performed a preliminary evaluation of the workflow with two knowledge engineers, working on novel mHealth app design tasks.

The CAPABLE project aims to improve the wellbeing of cancer patients managed at home via a mobile Coaching System recommending physical and mental health interventions. Patient reported outcomes are important for evaluation of the efficacy of these interventions. Nevertheless a large number of surveys might be overwhelming to patients. To understand the cognitive demand caused by the surveys and to find the adequate time to prompt patients to complete them we carried out a feasibility study. In this study we developed a machine learning cognitive load detector from blood volume pulse (BVP) captured by a photoplethysmography (PPG) signal. PPG sensors are available on consumer-grade smartwatches, which we will use in our Coaching System. We found that personalised 1D convolutional neural networks trained on raw BVP signal performed better in binary high vs low cognitive load classification than the personalised Support Vector Machines trained with heart rate variability and BVP features. We investigated if the further improvements can be obtained by teacher-student semi-supervised model training, nevertheless the performance gains were not notable. In the future we will include additional context information that might aid cognitive load estimation and drive both survey design as well as the timing of the prompts.

INTRODUCTION: Thanks to improvement of care, cancer has become a chronic condition. But due to the toxicity of treatment, the importance of supporting the quality of life (QoL) of cancer patients increases. Monitoring and managing QoL relies on data collected by the patient in his/her home environment, its integration, and its analysis, which supports personalization of cancer management recommendations. We review the state-of-the-art of computerized systems that employ AI and Data Science methods to monitor the health status and provide support to cancer patients managed at home. OBJECTIVE: Our main objective is to analyze the literature to identify open research challenges that a novel decision support system for cancer patients and clinicians will need to address, point to potential solutions, and provide a list of established best-practices to adopt. METHODS: We designed a review study, in compliance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, analyzing studies retrieved from PubMed related to monitoring cancer patients in their home environments via sensors and self-reporting: what data is collected, what are the techniques used to collect data, semantically integrate it, infer the patient's state from it and deliver coaching/behavior change interventions. RESULTS: Starting from an initial corpus of 819 unique articles, a total of 180 papers were considered in the full-text analysis and 109 were finally included in the review. Our findings are organized and presented in four main sub-topics consisting of data collection, data integration, predictive modeling and patient coaching. CONCLUSION: Development of modern decision support systems for cancer needs to utilize best practices like the use of validated electronic questionnaires for quality-of-life assessment, adoption of appropriate information modeling standards supplemented by terminologies/ontologies, adherence to FAIR data principles, external validation, stratification of patients in subgroups for better predictive modeling, and adoption of formal behavior change theories. Open research challenges include supporting emotional and social dimensions of well-being, including PROs in predictive modeling, and providing better customization of behavioral interventions for the specific population of cancer patients.

Operationalizing care delivery through an interdisciplinary healthcare team (IHT) requires knowledge about the overall structure of an IHT and the behavioral rules that "control" the dynamics of this structure interpreted as team and clinical leadership maintenance and task allocation. While progress has been made in understanding IHT structure, there is less work on the behavioral aspects of an IHT associated with its dynamics. In this paper we fill this void by extending our Team and Workflow Management Framework (TWMF) with a set of rules to operationalize IHT behavior in terms of clinical leadership, coordination of workflow execution over multiple days as part of continuity of care, and management of tasks, including urgent ones that prevent planned workflow execution. We briefly describe a proof-of-concept implementation of extended TWMF in the form of a computer system for supporting cooperative execution of clinical workflows by an IHT. The system is built on top of an existing business workflow execution engine and employs behavioral rules to control the IHT behavior. We also illustrate the operations of TWMF in a case study where an IHT is executing a workflow for the management of post-operative inpatient recovery after radical prostatectomy.

In the paper we describe a sequential classification scheme that iteratively explores levels of abstraction in the description of examples. These levels of abstraction represent attribute values of increasing precision. Specifically, we assume attribute values constitute an ontology (i.e., attribute value ontology) reflecting a domain-specific background knowledge, where more general values subsumes more precise ones. While there are approaches that consider levels of abstraction during learning, the novelty of our proposal consists in exploring levels of abstraction when classifying new examples. The described scheme is essential when tests that increase precision of example description are associated with costs – such a situation is often encountered in medical diagnosis. Experimental evaluation of the proposed classification scheme combined with ontological Bayes classifier (i.e., a näıve Bayes classifier expanded to handle attribute value ontologies) demonstrates that the classification accuracy obtained at higher levels of abstraction (i.e., more general description of classified examples) converges very quickly to the classification accuracy for classified examples represented precisely. This finding indicates we should be able to reduce the number of tests and thus limit their cost without deterioration of the prediction accuracy.

In the paper we present IIvotes – a new framework for constructing an ensemble of classifiers from imbalanced data. IIvotes incorporates the SPIDER method for selective data pre-processing into the adaptive Ivotes ensemble. Such an integration is aimed at improving balance between sensitivity and s pecificity (evaluated by the G-mean measure) for the minority class in comparison with single classifiers also combined with SPIDER. Using SPIDER to pre-process specific learning samples inside the ensemble improves sensitivity of derived component classifiers. At the same time the controlling mechanism of IIvotes ensures that overall accuracy (and thus specificity) is kept at a reasonable level. The new proposed IIvotes ensemble was thoroughly evaluated in a series of experiments where we tested it with symbolic (decision trees and rules) and non-symbolic (Naive Bayes) component classifiers. The results confirmed that combining SPIDER with an ensemble improved the performance (in terms of the G-mean measures) in comparison to a single classifier with SPIDER for all tested types of classifiers and two SPIDER pre-processing options (weak and strong amplification). These advantages were especially evident for decision trees and rules where differences between single and ensemble classifiers with SPIDER were more significant for both pre-processing options than for Naive Bayes. Moreover, the results demonstrated advantages of using a special abstaining classification strategy inside IIvotes rule ensembles, where component rule-based classifiers may refrain from predicting a class when in doubt. Abstaining rule ensembles performed much better with regard to G-mean than their non-abstaining variants.