Identification of motor progression in Parkinson's disease using wearable sensors and machine learning.

Wearable devices offer the potential to track motor symptoms in neurological disorders. Kinematic data used together with machine learning algorithms can accurately identify people living with movement disorders and the severity of their motor symptoms. In this study we aimed to establish whether a combination of wearable sensor data and machine learning algorithms with automatic feature selection can estimate the clinical rating scale and whether it is possible to monitor the motor symptom progression longitudinally, for people with Parkinson's Disease. Seventy-four patients visited the lab seven times at 3-month intervals. Their walking (2-minutes) and postural sway (30-seconds,eyes-closed) were recorded using six Inertial Measurement Unit sensors. Simple linear regression and Random Forest algorithms were utilised together with different routines of automatic feature selection or factorisation, resulting in seven different machine learning algorithms to estimate the clinical rating scale (Movement Disorder Society- Unified Parkinson's Disease Rating Scale part III; MDS-UPDRS-III). Twenty-nine features were found to significantly progress with time at group level. The Random Forest model revealed the most accurate estimation of the MDS-UPDRS-III among the seven models. The model estimations detected a statistically significant progression of the motor symptoms within 15 months when compared to the first visit, whereas the MDS-UPDRS-III did not capture any change. Wearable sensors and machine learning can track the motor symptom progression in people with PD better than the conventionally used clinical rating scales. The methods described in this study can be utilised complimentary to the clinical rating scales to improve the diagnostic and prognostic accuracy.

Antiparkinsonian medication masks motor signal progression in de novo patients.

Patients not yet receiving medication provide insight to drug-naïve early physiology of Parkinson's Disease (PD). Wearable sensors can measure changes in motor features before and after introduction of antiparkinsonian medication. We aimed to identify features of upper limb bradykinesia, postural stability, and gait that measurably progress in de novo PD patients prior to the start of medication, and determine whether these features remain sensitive to progression in the period after commencement of antiparkinsonian medication. Upper limb motion was measured using an inertial sensor worn on a finger, while postural stability and gait were recorded using an array of six wearable sensors. Patients were tested over nine visits at three monthly intervals. The timepoint of start of medication was noted. Three upper limb bradykinetic features (finger tapping speed, pronation supination speed, and pronation supination amplitude) and three gait features (gait speed, arm range of motion, duration of stance phase) were found to progress in unmedicated early-stage PD patients. In all features, progression was masked after the start of medication. Commencing antiparkinsonian medication is known to lead to masking of progression signals in clinical measures in de novo PD patients. In this study, we show that this effect is also observed with digital measures of bradykinetic and gait motor features.

Artificial intelligence outperforms human students in conducting neurosurgical audits.

OBJECTIVES: Neurosurgical audits are an important part of improving the safety, efficiency and quality of care but require considerable resources, time, and funding. To that end, the advent of the Artificial Intelligence-based algorithms offered a novel, more economically viable solution. The aim of the study was to evaluate whether the algorithm can indeed outperform humans in that task. PATIENTS & METHODS: Forty-six human students were invited to inspect the clinical notes of 45 medical outliers on a neurosurgical ward. The aim of the task was to produce a report containing a quantitative analysis of the scale of the problem (e.g. time to discharge) and a qualitative list of suggestions on how to improve the patient flow, quality of care, and healthcare costs. The Artificial Intelligence-based Frideswide algorithm (FwA) was used to analyse the same dataset. RESULTS: The FwA produced 44 recommendations whilst human students reported an average of 3.89. The mean time to deliver the final report was 5.80 s for the FwA and 10.21 days for humans. The mean relative error for factual inaccuracy for humans was 14.75 % for total waiting times and 81.06 % for times between investigations. The report produced by the FwA was entirely factually correct. 13 out of 46 students submitted an unfinished audit, 3 out of 46 made an overdue submission. Thematic analysis revealed numerous internal contradictions of the recommendations given by human students. CONCLUSION: The AI-based algorithm can produce significantly more recommendations in shorter time. The audits conducted by the AI are more factually accurate (0 % error rate) and logically consistent (no thematic contradictions). This study shows that the algorithm can produce reliable neurosurgical audits for a fraction of the resources required to conduct it by human means.

Diagnostic accuracy of frontotemporal dementia. An artificial intelligence-powered study of symptoms, imaging and clinical judgement.

PURPOSE: Frontotemporal dementia (FTD) is a neurodegenerative disorder associated with a poor prognosis and a substantial reduction in quality of life. The rate of misdiagnosis of FTD is very high, with patients often waiting for years without a firm diagnosis. This study investigates the current state of the misdiagnosis of FTD using a novel artificial intelligence-based algorithm. PATIENTS & METHODS: An artificial intelligence algorithm has been developed to retrospectively analyse the patient journeys of 47 individuals diagnosed with FTD (age range 52-80). The algorithm analysed the efficiency of patient pathways by utilizing a reward signal of ‒1 to +1 to assess the symptoms, imaging techniques, and clinical judgement in both behavioural and language variants of the disease. RESULTS: On average, every patient was subjected to 4.93 investigations, of which 67.4% were radiological scans. From first presentation it took on average 939 days for a firm diagnosis. The mean time between appointments was 204 days, and the average patient had their diagnosis altered 7.37 times during their journey. The algorithm proposed improvements by evaluating the interventions that resulted in a decreased reward signal to both the individual and the population as a whole. CONCLUSIONS: The study proves that the algorithm can efficiently guide clinical practice and improve the accuracy of the diagnosis of FTD whilst making the process of auditing faster and more economically viable.