.jpg)
Home-based physiotherapy increasingly benefits from inertial sensors (IMUs) that record patient movement. However, the lack of professional supervision leads to incorrect executions and the need for automatic evaluation systems. A key challenge is ambiguity: many repetitions fall on the boundaries between categories, where even experts disagree. Traditional classifiers, trained with rigid labels, ignore this uncertainty and assign a single class, losing valuable information.
Recent research proposes representing the complete label distribution for each repetition, rather than a single one. By minimizing the Kullback-Leibler divergence, the model learns not only the most likely category but also the degree of ambiguity. This allows automatically detecting whether a movement is ambiguous and which classes are relevant, improving reliability without losing classification accuracy.
This approach has direct applications in digital health platforms. Companies like Q2BSTUDIO, specialized in custom applications, can incorporate these techniques into remote rehabilitation systems. Integrating artificial intelligence with IMU sensors enables creating solutions that not only classify but also report on movement quality and ambiguity, facilitating informed clinical decisions.
At Q2BSTUDIO we develop custom software for the healthcare sector, using AWS and Azure cloud services to scale real-time analysis. We implement cybersecurity measures to protect sensitive patient data. Additionally, we offer business intelligence services with Power BI to visualize performance and ambiguity metrics. We also build AI agents that interact with patients and adjust exercise plans based on detected uncertainty.
The ability to represent ambiguity is not limited to IMUs; it is relevant in medical diagnostics, industrial quality control, or signal analysis. Companies looking for AI for business can adopt these models to gain robustness and transparency. At Q2BSTUDIO we help design machine learning systems that incorporate uncertainty, improving trust in predictions. If your organization needs to integrate movement analysis or classification with ambiguity, we invite you to explore our customized artificial intelligence solutions.
