OBJECTIVES: The increasing popularities of traditional Chinese medicine (TCM) and telehealth indicate a need for digitalizing major clinical assessment methods used during TCM consultations. In this study, an electronic TCM pulse diagnostic system was developed, and its validity was explored. MATERIALS AND METHODS: The system was developed with an artificial neural network (ANN). The output neurons were TCM pulse qualities operationalized as the intensity of eight elements (depth, rate, regularity, width, length, smoothness, stiffness, and strength) at six locations (left and right cun, guan, and chi). The input neurons were physical parameters of arterial pressure waveform acquired from the six locations by a pulse acquisition device. TCM pulse quality was rated by a TCM doctor on a 0-10 visual analog scale. Physical parameters were extracted from the arterial pressure waveform with a pulse extraction program developed in-house. The model structure, including number of hidden neurons and hidden layers, and training algorithms were manipulated to optimize model performance. The value of r² was the outcome measure indicating model performance. RESULTS: Two hundred twenty-nine subjects were recruited. Four-layer ANN models trained with 45 hidden neurons and the Levenberg-Marquardt algorithm performed the best. The r² ranged from 0.60 to 0.86. CONCLUSIONS: The validity of the proposed system generated by ANN is established and can assist TCM doctors in collecting relevant health data during telehealth consultation.
OBJECTIVES: The increasing popularities of traditional Chinese medicine (TCM) and telehealth indicate a need for digitalizing major clinical assessment methods used during TCM consultations. In this study, an electronic TCM pulse diagnostic system was developed, and its validity was explored. MATERIALS AND METHODS: The system was developed with an artificial neural network (ANN). The output neurons were TCM pulse qualities operationalized as the intensity of eight elements (depth, rate, regularity, width, length, smoothness, stiffness, and strength) at six locations (left and right cun, guan, and chi). The input neurons were physical parameters of arterial pressure waveform acquired from the six locations by a pulse acquisition device. TCM pulse quality was rated by a TCM doctor on a 0-10 visual analog scale. Physical parameters were extracted from the arterial pressure waveform with a pulse extraction program developed in-house. The model structure, including number of hidden neurons and hidden layers, and training algorithms were manipulated to optimize model performance. The value of r² was the outcome measure indicating model performance. RESULTS: Two hundred twenty-nine subjects were recruited. Four-layer ANN models trained with 45 hidden neurons and the Levenberg-Marquardt algorithm performed the best. The r² ranged from 0.60 to 0.86. CONCLUSIONS: The validity of the proposed system generated by ANN is established and can assist TCM doctors in collecting relevant health data during telehealth consultation.
Authors: Maximilian Moser; Matthias Frühwirth; Dietmar Messerschmidt; Nandu Goswami; Leopold Dorfer; Frank Bahr; Gerhard Opitz Journal: Front Physiol Date: 2017-07-21 Impact factor: 4.566