2025, 52(4):124-134.
Abstract:Due to the characteristics of electronic medical records (EMRs), such as the diversity of data types and temporal irregularity inherent, most existing deep learning-based methods cannot simultaneously capture static correlations between different types of clinical data and dynamic temporal dependencies between visits during the feature learning process. To address this issue, this paper proposes a disease prediction model based on multi-domain graph neural network. In this model, a temporal feature learning module that combines code level attention and time aware LSTM is first utilized to obtain the initial feature representation of patient visits. Then, based on the correlation and time interval information between different visits, a visit affinity graph and a visit sequence graph are constructed, and a graph convolutional neural network is used to mine the static and dynamic semantic associations between visit records from these graphs. Finally, a multi-domain feature fusion module based on self-attention mechanism is utilized to combine temporal features and semantic association features to obtain the final patient fusion representation for future disease prediction. The experimental results on two real clinical datasets show that our method outperforms other existing methods and achieves higher prediction accuracy.
2020, 47(4):180-188.
Abstract:A software-defined network (SDN) widely used in hospitals is difficult to deal with internal network threats. A Bayesian-based trust management mechanism is designed to identify possible malicious devices inside the network. The mechanism mainly uses the Bayesian inference method to derive the probability of sending malicious attack packets to realize the trust management of the internal devices of the network. Experiments in the simulation environment and the real network environment prove that the method can reduce the trust value of malicious devices faster than similar methods..
2016, 43(2):141-149.
Abstract:In order to solve the problem of the high accuracy delineation of biological target volume (BTV) for the radiotherapy of head and neck cancer, a random walk method was proposed by using PET (positron emission computed tomography) image features of tumors.Firstly, the selected region of interest (ROI) was segmented into the primary tumor (labeled as foreground seeds), normal tissue (labeled as background seeds) and pending region by three-dimensional adaptive region growing and morphological dilation based on PET SUV images.Secondly, due to the differences of contrast texture feature of head and neck tumor and surrounding normal tissues in PET images, the contrast texture feature was incorporated into the weights of random walk(RW) to further improve the accuracy of tumor segmentation results.Clinical PET image segmentations of head and neck cancer have shown that the improved RW is 9.34 times faster than the traditional RW on average.And the similarity is increased by 32.5% on average if the gross tumor volume delineated by clinicians is considered as the ground truth (P<0.05).The proposed method is an efficient and accurate method for the delineation of the BTV corresponding to head and neck tumors.
2006, 33(4).
Abstract:This paper applied Bayesian network to the syndrome differentiation research to improve the quantification of syndrome differentiation diagnosis system in traditional Chinese medicine(TCM).916 syndromes,51 key pattern elements and 1700 syndrome names as n