Application of Hierarchical Temporal Memory to Anomaly Detection of Vital Signs for Ambient Assisted Living

This thesis presents the development of a framework for anomaly detection of vital signs for an Ambient Assisted Living (AAL) health monitoring scenario. It is driven by spatiotemporal reasoning of vital signs that Cortical Learning Algorithms (CLA) based on Hierarchal Temporal Memory (HTM) theory undertakes in an AAL health monitoring scenario to detect anomalous data points preceding cardiac arrest.

This thesis begins with a literature review on the existing Ambient intelligence (AmI) paradigm, AAL technologies and anomaly detection algorithms used in a health monitoring scenario. The research revealed the significance of the temporal and spatial reasoning in the vital signs monitoring as the spatiotemporal patterns of vital signs provide a basis to detect irregularities in the health status of elderly people.

The HTM theory is yet to be adequately deployed in an AAL health monitoring scenario. Hence HTM theory, network and core operations of the CLA are explored. Despite the fact that standard implementation of the HTM theory comprises of a single-level hierarchy, multiple vital signs, specifically the correlation between them is not sufficiently considered. This insufficiency is of particular significance considering that vital signs are correlated in time and space, which are used in the health monitoring applications for diagnosis and prognosis tasks.

This research proposes a novel framework consisting of multi-level HTM networks. The lower level consists of four models allocated to the four vital signs, Systolic Blood Pressure (SBP), Diastolic Blood Pressure (DBP), Heart Rate (HR) and peripheral capillary oxygen saturation (SpO2) in order to learn the spatiotemporal patterns of each vital sign. Additionally, a higher level is introduced to learn spatiotemporal patterns of the anomalous data point detected from the four vital signs. The proposed hierarchical organisation improves the model’s performance by using the semantically improved representation of the sensed data because patterns learned at each level of the hierarchy are reused when combined in novel ways at higher levels.
To investigate and evaluate the performance of the proposed framework, several data selection techniques are studied, and accordingly, a total record of 247 elderly patients is extracted from the MIMIC-III clinical database.

The performance of the proposed framework is evaluated and compared against several state-of-the-art anomaly detection algorithms using both online and traditional metrics. The proposed framework achieved 83% NAB score which outperforms the HTM and k-NN algorithms by 15%, the HBOS and INFLO SVD by 16% and the k-NN PCA by 21% while the SVM scored 34%. The results prove that multiple HTM networks can achieve better performance when dealing with multi-dimensional data, i.e. data collected from more than one source/sensor.