Algorithmic complexity of EEG for neurodegenerative disease progression

As we have already discussed in a previous blog on Brain Consciousness and Complexity, algorithmic complexity is an intrinsic property of the brain dynamics, and be estimated through the Lempel-Ziv-Welch algorithm from the EEG (electroencephalogram) signals (Figure 1).

In this blog we discuss our recent paper on the relationship between algorithmic complexity and prognosis of neurodegeneration in idiopathic rapid eye movement behaviour disorder (RBD) (Ruffini et. al., 2018 accepted in Annals of Biomedical Engineering).

To begin with, RBD is a middle-aged or elderly male-predominant sleep disorder, characterised by vivid dreaming together with dream-enacting behaviours. The name of this disorder comes from the famous REM (rapid-eye-movement) sleep phase, which is associated with dreaming, and accounts for the 20-25% of the total sleep period in most adults. RBD is, thus, a disorder characterised by acting out dreams that are vivid, intense, and sometimes violent. Some of the symptoms of this disorder include talking, yelling, punching, kicking, and jumping from bed among others. The major complaint involves hurting oneself while sleeping. RBD is now considered to be an early stage of a-synucleinopathies, and can provide an early view of future brain health, as it can eventually evolve into other a-synucleinopathies, namely Parkinson’s disease (PD) and Dementia with Lewy Bodies (DLB).

One of the most studied aspects of EEG for a-synucleinopathy progression is the “slowing down” of EEG with disease progression, often measured by the “slow to fast” ratio that is the ratio of power in delta and theta bands over alpha and beta bands (Fantini et. al., 2003; Rodrigues-Brazete et. al., 2016). However, the healthy brain generates apparently complex data, and this complexity can be associated with cognitive health and conscious stage. A healthy brain engaging in modelling, prediction, and interaction with the world is expected to produce more complex-looking patterns, thus, in this work we expected that the complexity metrics (LZW, entropy, and mutual algorithmic information) would decrease with disease progression.

We thus studied 2-min data of 114 RBD patients, 19 of which developed PD after 1-10 years follow-up, 12 developed DLB, and 83 remained RBD. We also studied the data of 83 healthy controls (HC). As expected, our results revealed diminished complexity (LZW and entropy) of RBD compared to HC and even lower for the RBD ones that later developed PD or DLB. The mutual algorithmic information (MAI) index depends on a threshold for its computation. Analysing the results for a variety of thresholds we found a good discrimination across the different group categories (HC, RBD, PD, DLB) for many thresholds, indicating that the MAI can also right serve as a tool for prediction of RBD disease progression (Figure 2). The performance of the proposed complexity metrics is very well aligned with the slow-to-fast ratio, although not correlated with it, implying that information redundancies across channels, frequency bands, and time epochs can be exploited for compression and provide relevant insights for neurodenenerative disease progression. This conclusion was also highlighted in a further classification analysis, in which integrating all metrics together (entropy, LZW, MAI and slow to fast ratio) yielded excellent performance.

We are currently applying these metrics in a variety of EEG scenarios. Stay tuned to learn about new insights!

Leave a Reply

Your email address will not be published.