Research pick: Don’t go hacking my chart - "Anomaly detection architecture for smart hospitals based on machine learning, time series, and image recognition analysis: survey"

One of the big problems facing the use of digital systems in healthcare is the matter of security. Research in the International Journal of Medical Engineering and Informatics offers a new approach to defending against cyber-related patient-safety risks in the so-called smart hospital. The approach uses an anomaly-detection system that can analyse the full range of data generated by modern medical systems. By integrating numerical time-series analysis with image-based classification techniques it can identify irregularities that existing tools often miss.

Anomalies in this context are any unexpected deviations in a device’s behaviour or data stream, whether a sudden spike in a sensor reading, a breach of a device’s operating constraints, or an unusual pause in data transmission. While such events can indicate technical faults, they may also signal security breaches. Given that whole healthcare systems have been the subject of cyber-attack in recent years and suffered major outages as a result, there is a growing need for protection.

As hospitals begin to use more and more interconnected devices, such as monitors and wearable sensors, the vulnerabilities will only continue to grow. The researchers point out that even minor disruptions can cascade into clinical delays or expose systems to malicious interference.

The proposed system can manage the increasing complexity of electronic healthcare systems by using feature extraction to filter out the digital noise and highlight only the relevant relationships in the data.

One obstacle that is difficult to overcome is how to test and demonstrate the efficacy of the system, as there is a scarcity of real-life clinical datasets with which to work. The researchers plan to generate synthetic but representative datasets to evaluate each component of their detection architecture. They hope to develop it so that it can minimise false alarms while capturing irregularities in a timely manner. Their success will lead to security tools that could underpin digital healthcare as hospitals become ever more data-driven.

Haiba, S. and Mazri, T. (2025) ‘Anomaly detection architecture for smart hospitals based on machine learning, time series, and image recognition analysis: survey’, Int. J. Medical Engineering and Informatics, Vol. 17, No. 7, pp.1–14.