F181. Algorithm design for portable accelerometers for the detection, cuantification, and differentiation of the tremors in parkinson disease vs essential tremor in controlled and daily living conditions

Introduction The cardinal sign of essential tremor (ET) and Parkinson Disease (PD) is tremor, which is part of the diagnostic criteria. But, in early stages of the disease and in atypical presentations it is difficult to evaluate the tremors, the clinical examination has an extensive variability and it has been poorly studied. Accelerometers can quantify and differentiate the tremor, but, rarely have they been used in daily living conditions. The objective was to design a supervised learning algorithm that can detect, quantify, and differentiate the movements of a portable accelerometer in ET and PD during controlled and daily living conditions. Methods A descriptive and comparative clinical study. It enrolled 30 ET and 30 PD patients. The clinical and sociodemographic variables (age, sex, HY dominant frequency and harmonic. An algorithm was designed (rate 0.1 with a 1000 trees). Inferential statistics was used, and a normalized confusion matrix was calculated, sensitivity, specificity, and accuracy. Results Performance was classified in the normalized confusion matrix which allowed the distinction of tremor from no tremor in ET (86%) and PD (71%). The mathematical tree algorithm of decisions showed in: PD sensitivity: 65.4–69.8%, specificity: 72.8–90.9% and accuracy: 71.3–85.1%. In ET sensitivity: 77.6–93.8%, specificity 80.5–96.2% and accuracy: 79.5–95.6%. Being better in ET than in PD. Conclusion The tree algorithm of decisions in accelerometers allows the detection, quantification, and differentiation of the tremors of ET and PD during daily living conditions.