5 results about "External validation" patented technology

A double-module dynamic series cascade network system for predicting preoperative T stage of gastric cancer belongs to the technical field of medical artificial intelligence. The system adopts a deep learning architecture of double-module dynamic series connection. The first module realizes T1-T4 stage screening based on a hybrid model of parallel CNN and hierarchical Transformer. If it is judged as T1-T3 stage, the output result is output, and the second module is not entered. If it is judged as T4 stage, the second module is automatically triggered to perform T4 subtype differentiation task based on ResNet-152 submodel, and the output result is T4a or T4b. The system uses postoperative pathological results as the T stage gold standard, and shows high accuracy and universality in multicenter retrospective and prospective verification. The results show that the macro average AUC of the model in external verification reaches 0.964, the accuracy is 94.4%, and the T4 subtype recognition accuracy is highest, reaching 96.2%. The present application does not depend on labeled data, can significantly improve the accuracy and consistency of preoperative staging of gastric cancer, realize automatic and fine intelligent evaluation, has strong generalization ability and important clinical application value.

The invention relates to the technical field of Internet of Things, and discloses an Internet of Things-based equipment predictive maintenance method, which comprises the following steps: S1, constructing a multi-modal knowledge graph and a physical principle knowledge base, and integrating heterogeneous data modeling equipment knowledge; s2, detecting that equipment runs abnormally, and processing time sequence sensor data in real time to identify an abnormal signal; s3, judging an abnormal mode type; s4, performing dynamic causal reasoning, and if the fault is known, executing deterministic causal reasoning to output a fault causal chain; if the fault is an unknown fault, executing hypothetical causal path generation in combination with a physical principle knowledge base and an AI model; and S5, generating a self-adaptive maintenance work order. Through the knowledge graph self-correction step, the external verification feedback is received, the hypothetical causal path is solidified or removed, closed-loop self-evolution of knowledge is achieved, data-driven experience knowledge and principle-driven axiom knowledge are deeply fused, and the accuracy and interpretability of complex fault diagnosis are remarkably improved.

The invention belongs to the technical field of biological medicine, and particularly relates to a wide-period small cell lung cancer immune combined chemotherapy prognostic marker combination and application thereof. The biomarker combination is easy and convenient to sample, biopsy tissue does not need to be provided, and only peripheral blood needs to be taken for detection. The biomarker combination is high in accuracy, training is concentrated, the area under an AUC curve of a combined prediction model is 0.803, the optimal Cut-off value is 0.51, the sensitivity is 83.7%, and the specificity is 76.5%. The combined index prediction efficiency is good, and a single index cannot be predicted. In a constructed prediction model by applying independent external verification, the joint prediction AUC is 0.789, and when 0.51 is a Cut-off value, the sensitivity of the model is 82.1%, and the specificity is 63.6%. The combined index prediction efficiency is good, and a single index cannot be predicted. The method has practicability, and provides precise treatment guidance for clinical medicine.

The application provides a 2.5D medical image osteoporosis prediction method and system based on double fusion of a Transformer and stacking, and belongs to the field of artificial intelligence and medical image analysis. First, deep features are extracted in parallel from a 2.5D pelvic CT slice group by using a plurality of pre-training convolutional neural networks; then multi-head self-attention pre-fusion is performed by using a Transformer encoder to generate a unified image feature vector; then the unified image feature vector is spliced with key clinical features such as age and radiotherapy dose to form a multi-modal feature, and the multi-modal feature is input into a plurality of base classifiers to obtain preliminary prediction probabilities; finally, a stacking strategy is used for post-fusion to obtain a final osteoporosis risk probability. Experiments show that the method has excellent prediction performance, the external validation AUC reaches 0.964, and individualized bone damage risk assessment before radiotherapy can be realized.

The invention relates to the technical field of medicine, in particular to a column diagram model for predicting renal function prognosis of a CKD patient and an establishment method. The invention provides a column diagram model for predicting the probability that a CKD patient does not suffer from renal failure in 1, 2 and 3 years and an establishment method of the column diagram model, five risk factors including hypertension, mesangial hyperplasia degree, renal tubule atrophy degree, blood potassium and glomerular filtration rate are screened out from the column diagram model, and the five risk factors are used for establishing the column diagram model. In the training queue, the C index is 0.898, and the AUC values of the first year, the second year and the third year are 0.983 (95% CI 0.966-1.000), 0.955 (95% CI 0.924-0.986) and 0.945 (95% CI 0.918-0.972) respectively. In the external verification queue, the C index is 0.918, and the corresponding AUC values of the first year, the second year and the third year are 0.919 (0.853 to 0.986), 0.902 (0.822 to 0.982) and 0.915 (0.846 to 0.985). The calibration curve and the DCA curve show that the accuracy of the model is in a fitting range, which shows that the column graph model has good prediction capability.