Background and Purpose Adverse cosmesis or poor aesthetic appearance of the breast after breast conservation surgery (BCS) and/or radiation therapy (RT) is highly correlated with quality-of-life (QOL) in breast cancer patients. Existing methods for objective assessment of cosmesis utilize human-designed features that are calculated using outlines and/or key-points from breasts. These are complicated and time consuming to obtain. Our purpose is to develop and validate a deep neural network (DNN)-based approach for classifying adverse cosmesis from frontal digital images of patients who have undergone BCS and RT using an approximate region-of-interest (ROI) containing the breast. Materials and Methods We used 6 datasets (5 from BCS/RT-related clinical trials) to conduct our study. Training subsets (80% of patients) from 3 datasets (RAPID, OPAR, and PORTO) were used to train 3 independent support vector classifiers for classifying adverse cosmesis using the features extracted from preprocessed, approximate breast ROIs passed through a pretrained DNN, called ResNet-18. The trained classifiers/models were evaluated through 5 experiments: holdout validation, treatment-arm specific validation in RAPID, external validation (included datasets from the remaining 3 trials), assessment of the effect of preprocessing, and comparison with BCCT.core software. Area-under-the receiver operating characteristics (AUC) was used as the primary evaluation measure. Treatment-arm specific probability values were compared by the Mann-Whitney U test. Results AUC (0.78, 0.88, 0.94) values increased from RAPID to OPAR and PORTO respectively for holdout validation. The median probability of adverse cosmesis was higher in the same treatment arm in accordance with the interim results of the RAPID trial and differed significantly between the treatment arms (p = 0.006). For external validation, best AUC values were obtained by the model trained in RAPID (AUCs > 0.78). Preprocessing improved the performance to a large extent (average AUC: 0.87 vs 0.79 for internal validation, 0.78 vs 0.64 for external validation). AUCs of the individual models using DNN-based features were higher in 7 of 9 validation tests and statistically similar to the human-designed, dimensionless features extracted from BCCT.core software. Conclusion Appropriately preprocessed, approximate breast ROIs can be represented by DNN-based features to identify adverse cosmesis from frontal images of breast cancer patients treated with BCS/RT. Adaptations to this approach may be necessary for its applicability to images with non-frontal views and different quality. This work serves as a foundation for developing robust DNN-based platforms for evaluating cosmesis in clinical trials.

There are many design techniques to support the co-design of tangible technologies. However, few of these design methods allow the involvement of users at scale and across diverse geographic locations. While popular in psychology, the story completion method (SCM) has only recently started to be adopted within the HCI community. We explore whether SCM can generate meaningful design insights from large, diverse study populations for the design of Tangible User Interfaces (TUIs). Based on the results of two questionnaire studies using SCM, we conclude that the method can be used to generate meaningful design insights. Drawing on a systematic review of 870 TUI papers, we then contextualise the strengths and weaknesses of SCM against commonly used design methods, before reflecting on our experience of using the method across two distinct domains. We discuss the advantages of the method (particularly in terms of the scale and diversity of participation) and the challenges (particularly around constructing meaningful story stems, and developing the correct level of scaffolding to support creativity). We conclude that SCM is particularly suitable to be used in the early stages of the design process to understand the socio-cultural context of deployment.

Autonomous systems, such as drones and rescue robots, are increasingly used during emergencies. They deliver services and provide situational awareness that facilitate emergency management and response. To do so, they need to interact and cooperate with humans in their environment. Human behaviour is uncertain and complex, so it can be difficult to reason about it formally. In this article, we propose IDEA: an adaptive software architecture that enables cooperation between humans and autonomous systems, by leveraging the social identity approach. This approach establishes that group membership drives human behaviour. Identity and group membership are crucial during emergencies, as they influence cooperation among survivors. IDEA systems infer the social identity of surrounding humans, thereby establishing their group membership. By reasoning about groups, we limit the number of cooperation strategies the system needs to explore. IDEA systems select a strategy from the equilibrium analysis of game-theoretic models that represent interactions between group members and the IDEA system. We demonstrate our approach using a search-and-rescue scenario, in which an IDEA rescue robot optimises evacuation by collaborating with survivors. Using an empirically validated agent-based model, we show that the deployment of the IDEA system can reduce median evacuation time by 13.6%.

Resilience engineering (RE) is most commonly applied at the organisational level, and has historically been associated with safety-critical industries such as nuclear, medical or aviation. This paper explores the application of RE frameworks within software engineering, and investigates resilient performance of the socio-technical system that supports the creation of software. We present a preliminary study based on a secondary analysis of data from previous ethnographic studies of commercial software practice. This analysis uses an RE framework devised for small team practice in safety critical settings. We present and discuss three salient episodes of software practice that illustrate the application of RE principles to software engineering, and suggest how this kind of analysis may benefit software engineering. We present challenges and opportunities based on our experience and propose future research directions.

The pressure on software developers to produce secure software has never been greater. But what does security look like in environments that do not produce security-critical software? In answer to this question, this multi-sited ethnographic study characterizes security episodes and identifies five typical behaviors in software development. Using theory drawn from information security and motivation research in software engineering, this article characterizes key ways in which individual developers form security responses to meet the demands of particular circumstances, providing a framework managers and teams can use to recognize, understand, and alter security activity in their environments.

Despite the availability of various methods and tools to facilitate secure coding, developers continue to write code that contains common vulnerabilities. It is important to understand why technological advances do not sufficiently facilitate developers in writing secure code. To widen our understanding of developers' behaviour, we considered the complexity of the security decision space of developers using theory from cognitive and social psychology. Our interdisciplinary study reported in this article (1) draws on the psychology literature to provide conceptual underpinnings for three categories of impediments to achieving security goals, (2) reports on an in-depth meta-analysis of existing software security literature that identified a catalogue of factors that influence developers' security decisions, and (3) characterises the landscape of existing security interventions that are available to the developer during coding and identifies gaps. Collectively, these show that different forms of impediments to achieving security goals arise from different contributing factors. Interventions will be more effective where they reflect psychological factors more sensitively and marry technical sophistication, psychological frameworks, and usability. Our analysis suggests “adaptive security interventions” as a solution that responds to the changing security needs of individual developers and a present a proof-of-concept tool to substantiate our suggestion.

Peer-to-peer energy-trading platforms (P2P) have the potential to transform the current energy system. However, research is presently scarce on how people would like to participate in, and what would they expect to gain from, such platforms. We address this gap by exploring these questions in the context of the UK energy market. Using a qualitative interview study, we examine how 45 people with an interest in renewable energy understand P2P. We find that the prospective users value the collective benefits of P2P, and understand participation as a mechanism to support social, ecological and economic benefits for communities and larger groups. Drawing on the findings from the interview analysis, we explore broad design characteristics that a prospective P2P energy trading platform should provide to meet the expectations and concerns voiced by our study participants.

Research has established the wide variety of security failures in mobile apps, their consequences, and how app developers introduce or exacerbate them. What is not well known is why developers do so---what is the rationale underpinning the decisions they make which eventually strengthen or weaken app security? This is all the more complicated in modern app development's increasingly diverse demographic: growing numbers of independent, solo, or small team developers who do not have the organizational structures and support that larger software development houses enjoy.

The global population is ageing, leading to shifts in healthcare needs. Home healthcare monitoring systems currently focus on physical health, but there is an increasing recognition that psychological wellbeing also needs support. This raises the question of how to design devices that older adults can interact with to log their feelings. We designed three tangible prototypes, based on existing paper-based scales of affect. We report findings from a lab study in which participants used the prototypes to log the emotion from standardised emotional vignettes. We found that the prototypes allowed participants to accurately record identified emotions in a reasonable time. Our participants expressed a perceived need to record emotions, either to share with family/carers or for self-reflection. We conclude that our work demonstrates the potential for in-home tangible devices for recording the emotions of older adults to support wellbeing.

The world is facing an ageing population phenomenon, coupled with health and social problems, which affect older people’s ability to live independently. This situation challenges the viability of health and social services. Smart home technology can play a significant role in easing the pressure on caregivers, as well as reduce the financial costs of health and social services. Activity of Daily Living (ADL) recognition is an essential step to translate sensor data into activities at high semantic levels. Supervised Machine Learning (ML) algorithms are the most commonly used techniques for this application. However, a common problem is a lack of availability of enough annotated data to train these algorithms. Collecting annotated data is expensive, time consuming, and may violate people’s privacy. Intra- and inter-personal variation in performing complex activities is another challenge for an ML-based activity recognition approach. In this paper, a multi-layered knowledge-based architecture for recognising ADL in real-time is proposed. At the first stage, sensor data is pre-processed; events that describe changes in the environment are detected at the second stage, in which the sequence of events is used to recognise more semantically complex activities at the third stage. A new ADL ontology is proposed to model the knowledge related to the sensor platform and the targeted activities as the previously proposed ontologies were either designed to deal with specific sensor data, or they ignored the context environment information which is important in recognising complex activities.

Some online social networks (OSNs) allow users to define friendship-groups as reusable shortcuts for sharing information with multiple contacts. Posting exclusively to a friendship-group gives some privacy control, while supporting communication with (and within) this group. However, recipients of such posts may want to reuse content for their own social advantage, and can bypass existing controls by copy-pasting into a new post; this cross-posting poses privacy risks. This paper presents a learning to share approach that enables the incorporation of more nuanced privacy controls into OSNs. Specifically, we propose a reusable, adaptive software architecture that uses rigorous runtime analysis to help OSN users to make informed decisions about suitable audiences for their posts. This is achieved by supporting dynamic formation of recipient-groups that benefit social interactions while reducing privacy risks. We exemplify the use of our approach in the context of Facebook.

In this paper we show empirically that highly humanlike robots make thoughts of death more accessible, leading to perceptions of uncanniness and eeriness of such robots. Rather than reducing the humanlikeness of robots, our research suggests the addition of emotion displays to decrease a sense of uncanniness. We show that a highly humanlike robot displaying emotions in a social context reduces death-thought accessibility (DTA), which in turn reduces uncanniness. In a pre-test with N = 95 participants, we established that not all humanoid robots elicit thoughts of death and that the extent to which a robot appears humanlike may be linked to DTA. In our Main Study, N = 44 participants briefly interacted with a highly humanlike robotic head that either showed appropriate basic emotions or reacted by blinking. The display of emotions significantly reduced perceptions of uncanniness, which was mediated by a corresponding reduction in DTA. Implications for the design of humanoid robots are proposed.

In this paper, we report a breaching study to explore the use of a tele-operated humanoid avatar from both the perspective of the robot operator and the perspective of interlocutors. Humanoid robot avatars provide capabilities that video conferencing and mobile remote presence devices lack, particularly in multi-party conversations. Specifically, we were interested whether new users encountering a tele-operated humanoid avatar in a public setting would spontaneously use head movements and arm gestures to allow better conversational flow and group interaction. Video data shows that both operators and interlocutors initially explore the system's capabilities. Most operators used social gestures such as waving and pointing, as well as gaze which in turn led to related social responses from interlocutors. Interview data additionally show that robot operators and interlocutors rated the experience very positively. This paper discusses the implications of our findings for the design of `in the wild' experiments and the usability of tele-operated humanoid avatars.