.jpg)
In the field of physical rehabilitation, data from inertial sensors (IMU) allows for monitoring movement with astonishing precision. However, when artificial intelligence models are applied to classify gait patterns or detect anomalies, a critical problem arises: the explanations these models offer are often difficult for clinicians to interpret. While a physiotherapist thinks in terms of muscle groups or joint segments, traditional explainability methods work at the level of individual channels, generating scattered and biomechanically incoherent recommendations. This disconnect between machine logic and human reasoning limits the adoption of AI for healthcare companies, which need tools that truly support decision-making.
To overcome this barrier, approaches such as group counterfactual explanations emerge, which simultaneously modify sets of relevant features —for example, the activation of an entire muscle group— instead of altering isolated data points. This type of technique not only improves interpretability but also generates actionable recommendations: 'for your movement to be classified as correct, you must increase quadriceps activation and reduce hip rotation.' The key lies in achieving a balance between the validity of the counterfactual and the semantic cohesion of the groups. In this sense, proposals such as Learnable Gates methods demonstrate that it is possible to maintain the accuracy of the original model while imposing a group-based relevance assignment, something that Shapley-based methods cannot achieve on their own.
For companies developing technology-assisted rehabilitation solutions, implementing this type of architecture requires a comprehensive approach. From capturing and processing IMU signals to deployment in cloud environments, having a technology partner that offers AI for businesses is essential to transform research into clinically useful products. Furthermore, integrating these models with Power BI dashboards allows therapists to visualize suggested corrections in real time, while the use of AWS and Azure cloud services ensures scalability and security of patient data.
At Q2BSTUDIO, we develop custom applications that connect artificial intelligence with expert reasoning. Our business intelligence and custom software development services allow us to create systems that not only classify movements but also explain the reasoning behind each decision in terms that professionals understand. Additionally, we offer cybersecurity solutions to protect sensitive information throughout the entire data lifecycle, from sensor capture to cloud analysis. If your organization needs to build AI agents capable of generating coherent counterfactual explanations for rehabilitation or other domains, we can help you design and implement the necessary technological infrastructure.
