Deep learning for acupuncture point selection patterns based on veteran doctor experience of Chinese medicine

The inheritance of clinical experience of veteran doctors of Chinese medicine (CM) plays a key role in the development and effectiveness enhancement of Chinese medicine in the history. The clinical experience are classified as the patterns of disease diagnosis and Chinese medical Zheng diagnosis, the identification of core elements of Zheng, the treatment experience and relation of herbal medicine formulae, Zheng and disease, and the common law of diagnosis and treatment in real practice. The source of the experience mainly originates from literature and manuscripts of CM masters, which are being electronically recorded during the last two decades. As a result, it makes feasible to apply data mining to the knowledge discovery through the experience of veteran CM doctors. However, the current focus on this field is limited to the published literature such as journal papers, conference proceedings and textbooks, but the paper based manuscripts personally written by the veteran doctors are usually neglected. In this paper, we established a database for Dr Situ Ling, who is a deceased famous CM acupuncture master in southern China. The study objective is to discover the acupuncture point selection patterns which require profession knowledge and experience from senior CM doctors. It is believed these patterns are deposited as underlying knowledge with various middle level concepts that can be analyzed and discover by a serial of algorithms. Thus in this work, we formularized the patterns of acupuncture point selection as a learning task with deep architecture, which attempts to capture either existent or underlying concepts so as to simulate the planning process of the combined diagnosis of western medicine and Chinese medicine. The Restricted Boltzmann Machines (RBM) was used as the main model for deep learning to process to medical record data with international standard diagnosis (ICD-10) previously made by trained doctors. Then the ICD-10 based diagnosis dataset was introduced into our framework to enhance the concepts diversity. After applying this model, the learning accuracy based on the medical record database of Dr Situ Ling was raised up to 75%. Thus this model can serve as a solution to discover the acupuncture point selection patterns of CM acupuncture veteran doctors. Furthermore, the data mining study model linked by international diagnosis standard (i.e. ICD-10), point selection patterns, and clinical symptoms will provide useful cues to reveal the essence of Zheng diagnosis through experience of CM veteran doctors.