04 September 2026
Bilbao
The use of increasingly complex and accurate machine learning algorithms and tools in the healthcare sector is making it ever more difficult for doctors and psychologists to understand and trust the reasoning behind a prediction or diagnosis. To overcome this barrier of opacity, known as the ‘black box’, the thesis ‘Human-Centred Machine Learning for Health and Cognitive Modelling’, defended at the University of Deusto by Amaia Pikatza Huerga, proposes the use of ‘explainable AI’.
This methodology combines the analytical precision of advanced models with the interpretability required to ensure that medical and psychology professionals do not have to blindly validate the algorithm’s reasoning when making decisions. The author of the thesis proposes transparent, multimodal algorithms capable of explaining their decisions in the fields of medicine, psychology and cognitive assessment.
Thus, by ‘opening the black box’ of Artificial Intelligence, healthcare professionals can understand why the model predicts a patient’s readmission or relapse, turning artificial intelligence into a genuinely collaborative and reliable tool.
To demonstrate that this methodology works in real-world settings, the author of the thesis has tested it in four areas of medicine and psychology. In the field of heart failure, the complex algorithmic models developed by Pikatza were able to predict which specific patients would be readmitted to hospital within 30 days of discharge, with up to 30% greater accuracy than traditional statistical techniques. In addition to the predictive data, and thanks to the use of explainable AI, the researcher provided doctors with the specific reasons behind each prediction, taking into account factors such as frailty, anxiety levels and the patient’s clinical history.
In the case of patients with multiple chronic conditions, the author of the thesis demonstrated that incorporating social determinants of health makes it possible to predict more accurately which patients will be readmitted to hospital within 30 days of discharge. Thanks to explainable AI, in addition to improving the accuracy of the prediction, doctors were given an explanation showing that patients’ social environment and perceived well-being are decisive factors in relapse.
In the field of eating disorders, the researcher applied her algorithms to predict, one year ahead, both the risk of developing the disorder and each patient’s likelihood of recovery. Using explainable AI, she was able to demonstrate to specialists the decisive role that psychological factors such as resilience and perceived quality of life play in treatment outcomes, providing a personalised prognosis to guide clinical follow-up.
The fourth area studied in this thesis is human creativity, to which Amaia Pikatza has applied her models to automatically and objectively assess dimensions of divergent thinking, such as originality, flexibility and elaboration, based on drawings and texts created by people. In this case, explainable AI demonstrated the effectiveness of combining visual and textual information.
Transparency and ethical rigour
The research demonstrates that the technical performance of algorithms and their interpretability are not mutually exclusive objectives. By embedding the explanation within the algorithm, specialists are provided with a transparent overview of what lies behind each prediction.
This approach responds to the growing demand in Europe for the development of safe, ethical AI tools centred on human needs. By enabling healthcare and psychology professionals to verify algorithmic criteria against real clinical and cognitive data, the thesis lays the foundations for the future implementation of care systems that support decision-making without replacing expert judgement.
