Work number - M 6 AWARDED
Presented Presented by Kharkiv National University of Radio Electronics
Dr. Sc. Perova I.G.
The paper solves the scientific and technological problem of developing methods for analyzing the medical data streams in a sequential mode using Medical Data Mining approaches to support the implementation of eHealth in Ukraine and increase the efficiency of medical diagnosis in general. The mathematical software provides an opportunity to fill in the missing values in the patient's medical records in a sequential mode for the most effective use of all measured features, to assess the informative value of each of the features to justify the need for their further use or removal, to perform diagnostics in a sequential online mode with the calculation of the fuzzy membership function of each patient to each diagnosis. Based on the analysis of approaches, methods and systems of diagnostics and preprocessing of medical data, the low efficiency of their work with such data streams was revealed. So, it is necessary to combine (hybridize) several types of data analysis systems into one and develop methods of learning such systems taking into account the peculiarities of primary medical data.
Improved method of spatial extrapolation of diagnostic features of the patient through the use of fuzzy level of membership in the mode of filling in the missing values. The development of a hybrid method for assessing the informativeness of medical indicators by supplementing the Oja’s neuron with a contour of adaptive tuning makes it possible to reduce information processing time and increase the speed of the neural network self-learning process to identify informative features. Methods of analysis of medical data during medical online diagnosis in the controlled training mode on the basis of hybrid adaptive neuro-fuzzy systems for situations of representative educational sample are developed. A method of medical data analysis during medical online diagnosis in active learning mode based on hybrid neuro-fuzzy system automatically switches between controlled learning and self-learning modes and provides appropriate changes in the process of adjusting weights of neuro-fuzzy network according to individual patient characteristics. Proposed method helps to reduce data volumes in the training sample and network setup time. A method of medical data analysis during online medical diagnosis in self-learning mode, which is based on adaptive robust fuzzy clustering of multidimensional patient space using Manhattan metrics as the basis of fuzzy belonging, has developed a solution to clustering problems in the context of medical data distortion. A method of medical data analysis during medical online diagnosis in self-learning mode, which is based on neuro-fuzzy auto-associative memory, which consistently refines its parameters when receiving new medical data and automatically implements associative fuzzy inference based on analysis of precedents.
All developed methods were tested on clinical medical data, which was confirmed by the acts of implementation and demonstrated a significant increasing of accuracy and time of data processing.
Number of publications: 34, incl. 19 articles (8 – in English-language journals with an impact factor). The total number of references to the authors' publications / h-index of work, according to the databases is respectively: Web of Science – 16 / 3, Scopus – 62 / 6, Google Scholar – 139 / 8. Received 1 patents of Ukraine for inventions. 1 doctoral dissertation was defended on this topic.