Depression is a common and often recurring mental health condition. Many people who experience depression go through periods of varying depression symptoms. Traditional ways to monitor depression focus on questionnaires and interviews during clinic visits. These methods are useful but have clear limits. They rely on memory, give only a snapshot of someone’s condition, and may miss small but important changes in how a person feels or behaves outside of clinical assessments.

Today, wearable devices such as smartwatches and smartphones make it possible to track behaviours continuously in daily life. These behaviours include how much someone moves, how they sleep, and how their heart rate changes. These measurements are called digital biomarkers. They can be collected with minimal effort from the person and provide an objective picture of what happens in their body and behaviour from day to day. This method of using digital devices to measure behaviour and physical signals in real life is called digital phenotyping. It offers a new way to observe how people function outside the clinic, in their everyday environment, and helps detect subtle changes that may not be visible through traditional interviews or questionnaires.

This thesis explores how digital biomarkers interact with each other over time in individuals with varying levels of depression symptoms. The main idea is based on a network approach. In this approach, depression is not seen as one fixed condition, but as a system where different behaviours and symptoms can influence each other. A network illustrates which signals are connected and how changes in one signal can lead to changes in others. This makes it possible to see not only what changes, but how change happens....

A considerable portion of the analyses is presented in a confidential appendix, which is not publicly accessible

Introduction
Cancer-related pain is a prevalent and under-assessed issue in oncological care. Despite the availability of effective pain treatments, challenges in pain assessment - such as communication barriers, subjective interpretation, and workflow constraints - continue to hinder accurate pain recognition and management. In recent years, advances in artificial intelligence (AI), sensor technology and mobile health have offered promising new opportunities for more consistent, continuous and context-aware assessment strategies. Facial expressions and vocal cues have been identified as promising behavioural indicators of pain, and their use in automatic pain assessment (APA) is gaining increasing attention.

To explore how such technologies could support cancer pain assessment and management, the SENSAI project was initiated. The project focuses on the development of a human-centred, AI-empowered tool for APA in cancer care, based on facial expressions and vocal cues captured through a mobile application.

This thesis presents the initial steps of the SENSAI project. Its objectives were to examine current pain assessment practices and challenges in oncology, investigate end-user perspectives on AI-supported assessment, translate research insights into a conceptual tool framework, support the design of a user-centred APA application, and initiate the development of a multimodal cancer-related pain database.

Methodology
A multiphase, design-oriented research approach was used, guided by user-centred design, the Double Diamond framework, and a translational AI development model for healthcare. This thesis encompassed the Research and Conceptualisation phases and initiated the Development phase.

In the research phase, background studies and an exploratory interview study with oncologists were conducted to understand current pain assessment practices and challenges. This informed the problem definition and supported the development of an initial concept for the tool In the conceptualisation phase, this concept was refined into was structured framework. The second part of the interview study explored oncologists’ attitudes towards the proposed APA tool, guided by the mobile health Technology Acceptance Model. These findings, along with expert consultations, a brainstorm session and feasibility considerations, shaped the envisioned design and functionalities of the tool. In the development phase, agile application development was initiated in collaboration with software company Innovattic, with the author serving as product owner. In parallel, a clinical study protocol was prepared for the creation of a multimodal cancer-related pain dataset to support future AI model training.

Results
The interview study identified five key challenges in current cancer pain assessment: the complexity and subjectivity of pain, ambiguity in responsibility, communication barriers, balancing all pain information with clinical judgement to come to decision-making, and practical constraints. Oncologists expressed conditional interest in an APA tool, emphasising the importance of clinical validation, interpretability, and seamless integration. The conceptual framework proposed three core functionalities: multimodal data collection via a mobile application, AI-based pain classification, and feedback for both patients and clinicians. The mobile application is currently under development and includes measurement and login/authentication modules. A clinical protocol for database development has been submitted for ethical review.

Conclusion
This thesis establishes a solid interdisciplinary foundation for the development of a human-centred APA tool tailored to cancer care. While the AI model itself lies beyond the scope of this thesis, the design process, stakeholder engagement, and technical groundwork contribute to a clinically grounded and ethically aligned vision for future development. The work advances efforts to improve how cancer-related pain is recognised, communicated, and managed in clinical practice. Building on this foundation, future research should focus on user testing with patients, iterative refinement of the application and measurement protocol, technical development of the AI model, and broader clinical evaluation to assess its impact on care delivery and patient experience.