Explainable artificial intelligence for searching frequency characteristics in Parkinson's disease tremor.

The distinction between Parkinson's disease (PD) and essential tremor (ET) tremors is subtle, posing challenges in differentiation. To accurately classify the PD and ET, BiLSTM-based recurrent neural networks are employed to classify between normal patients (N), PD patients, and ET patients using accelerometry data on their lower arm (L), hand (H), and upper arm (U) as inputs. The trained recurrent neural network (RNN) has reached 80% accuracy. The neural network is analyzed using layer-wise relevance propagation (LRP) to understand the internal workings of the neural network. A novel explainable AI method, called LRP-based approximate linear weights (ALW), is introduced to identify the similarities in relevance when assigning the class scores in the neural network. The ALW functions as a 2D kernel that linearly transforms the input data directly into the class scores, which significantly reduces the complexity of analyzing the neural network. This new classification method reconstructs the neural network's original function, achieving a 73% PD and ET tremor classification accuracy. By analyzing the ALWs, the correlation between each input and the class can also be determined. Then, the differentiating features can be subsequently identified. Since the input is preprocessed using short-time Fourier transform (STFT), the differences between the magnitude of tremor frequencies ranging from 3 to 30 Hz in the mean N, PD, and ET subjects are successfully identified. Aside from matching the current medical knowledge on frequency content in the tremors, the differentiating features also provide insights about frequency contents in the tremors in other frequency bands and body parts.

Motion characteristics of subclinical tremors in Parkinson's disease and normal subjects.

The characteristics of the Parkinson's disease tremor reported previously are not applicable to the full spectrum of severity. The characteristics of high- and low-amplitude tremors differ in signal regularity and frequency dispersion, a phenomenon that indicates characterisation should be studied separately based on the severity. The subclinical tremor of Parkinson's disease is close to physiological tremor in terms of amplitude and frequency, and their distinctive features are still undetermined. We aimed to determine joint motion characteristics that are unique to subclinical Parkinson's disease tremors. The tremors were characterised by four hand-arm motions based on displacement and peak frequencies. The rest and postural tremors of 63 patients with Parkinson's disease and 62 normal subjects were measured with inertial sensors. The baseline was established from normal tremors, and the joint motions were compared within and between the two subject groups. Displacement analysis showed that pronation-supination and wrist abduction-adduction are the most and least predominant tremor motions for both Parkinson's disease and normal tremors, respectively. However, the subclinical Parkinson's disease tremor has significant greater amplitude and peak frequency in specific predominant motions compared with the normal tremor. The flexion-extension of normal postural tremor increases in frequency from the proximal to distal segment, a phenomenon that is explainable by mechanical oscillation. This characteristic is also observed in patients with Parkinson's disease but with amplification in wrist and elbow joints. The contributed distinctive characteristics of subclinical tremors provide clues on the physiological manifestation that is a result of the neuromuscular mechanism of Parkinson's disease.

Biomechanical System Versus Observational Rating Scale for Parkinson's Disease Tremor Assessment.

There is a lack of evidence that either conventional observational rating scale or biomechanical system is a better tremor assessment tool. This work focuses on comparing a biomechanical system and the Movement Disorder Society-sponsored revision of the Unified Parkinson's Disease Rating Scale in terms of test-retest reliability. The Parkinson's disease tremors were quantified by biomechanical system in joint angular displacement and predicted rating, as well as assessed by three raters using observational ratings. Qualitative comparisons of the validity and function are made also. The observational rating captures the overall severity of body parts, whereas the biomechanical system provides motion- and joint-specific tremor severity. The tremor readings of the biomechanical system were previously validated against encoders' readings and doctors' ratings; the observational ratings were validated with previous ratings on assessing the disease and combined motor symptoms rather than on tremor specifically. Analyses show that the predicted rating is significantly more reliable than the average clinical ratings by three raters. The comparison work removes some of the inconsistent impressions of the tools and serves as guideline for selecting a tool that can improve tremor assessment. Nevertheless, further work is required to consider more variabilities that influence the overall judgement.

An In-Laboratory Validity and Reliability Tested System for Quantifying Hand-Arm Tremor in Motions.

Despite the advancement of the tremor assessment systems, the current technology still lacks a method that can objectively characterize tremors in relative segmental movements. This paper presents a measurement system, which quantifies multi-degrees-of-freedom coupled relative motions of hand-arm tremor, in terms of joint angular displacement. In-laboratory validity and reliability tests of the system algorithm to provide joint angular displacement was carried out by using the two-degrees-of-freedom tremor simulator with incremental rotary encoder systems installed. The statistical analyses show that the developed system has high validity results and comparable reliability performances using the rotary encoder system as the reference. In the clinical trials, the system was tested on 38 Parkinson's disease patients. The system readings were correlated with the observational tremor ratings of six trained medical doctors. The moderate to very high clinical correlations of the system readings in measuring rest, postural and task-specific tremors add merits to the degree of readiness of the developed tremor measurement system in a routine clinical setting and/or intervention trial for tremor amelioration.