ObjectiveThis study aims to develop and validate a novel multimodal predictive model， termed criss-cross network（CCNet）-directed graph neural network（DGNN）（CGN）， for accurate assessment of pulmonary nodule progression in high-risk individuals for lung cancer， by integrating longitudinal chest computed tomography（CT） imaging with both traditional Chinese and western clinical evaluation data. MethodsA cohort of 4 432 patients with pulmonary nodules was retrospectively analyzed. A twin CCNet was employed to extract spatiotemporal representations from paired sequential CT scans. Structured clinical assessment and imaging-derived features were encoded via a multilayer perceptron， and a similarity-based alignment strategy was adopted to harmonize multimodal imaging features across temporal dimensions. Subsequently， a DGNN was constructed to integrate heterogeneous features， where nodes represented modality-specific embeddings and edges denoted inter-modal information flow. Finally， model optimization was performed using a joint loss function combining cross-entropy and cosine similarity loss， facilitating robust classification of nodule progression status. ResultsThe proposed CGN model demonstrated superior predictive performance on the held-out test set， achieving an area under the receiver operating characteristic curve（AUC） of 0.830， accuracy of 0.843， sensitivity of 0.657， specificity of 0.712， Cohen's Kappa of 0.417， and F₁ score of 0.544. Compared with unimodal baselines， the CGN model yielded a 36%-48% relative improvement in AUC. Ablation studies revealed a 2%-22% increase in AUC when compared to simplified architectures lacking key components， substantiating the efficacy of the proposed multimodal fusion strategy and modular design. Incorporation of traditional Chinese medicine （TCM）-specific symptomatology led to an additional 5% improvement in AUC， underscoring the complementary value of integrating TCM and western clinical data. Through gradient-weighted activation mapping visualization analysis， it was found that the model's attention predominantly focused on nodule regions and effectively captured dynamic associations between clinical data and imaging-derived features. ConclusionThe CGN model， by synergistically combining cross-attention encoding with directed graph-based feature integration， enables effective alignment and fusion of heterogeneous multimodal data. The incorporation of both TCM and western clinical information facilitates complementary feature enrichment， thereby enhancing predictive accuracy for pulmonary nodule progression. This approach holds significant potential for supporting intelligent risk stratification and personalized surveillance strategies in lung cancer prevention.

AIM:To evaluate the characteristics of ocular biometric parameters and the distribution of corneal astigmatism(CA)in patients with high myopia before cataract surgery.METHODS:A prospective cross-sectional study was conducted, and 695 cataract patients(695 eyes)with high myopia [defined as an axial length(AL)≥26.00 mm] scheduled to undergo cataract surgery at our hospital from January 2022 to December 2024 were consecutively enrolled, another 695 cataract patients(695 eyes)with normal ALs(22.00 mm ≤AL≤25.00 mm)who underwent cataract surgery at our hospital during the same period were included in the control group. For patients with both eyes eligible, the right eye was used for analysis. Before cataract surgery, IOL Master 700 was used to measure the ocular biometric parameters of both eyes for each patient in the two groups. The medical records and ocular biometric data in the two groups were recorded and collected.RESULTS:There were no statistically significant differences between the two groups in genger, age, corneal diameter, and central corneal thickness(all P>0.05). In the high myopia group, the mean AL was 29.20±2.61 mm, and 252 eyes(34.1%)had AL ≥30.00 mm(extremely high myopia). The mean anterior chamber depth(ACD), lens thickness, vitreous chamber depth(VCD), CA, AL/corneal radius of curvature and VCD/AL in the high myopia group were 3.45±0.40, 4.41±0.47, 21.34±2.60 mm, 1.18±0.78 D, 3.79±0.38, and 0.73±0.03, respectively, which were all greater than those in the control group(all P<0.01). In the high myopia group, 350 eyes(50.4%)had CA ≥1.00 D, 192 eyes(27.6%)had CA ≥1.50 D, and 94 eyes(13.5%)had CA ≥2.00 D, which were all higher than those in the control group(32.8%, 15.1%, and 6.6%, respectively; all P<0.001). In the high myopia group, 87 eyes(12.5%)had flat corneas, 424 eyes(61.0%)had moderate CA, and 40 eyes(5.8%)had high CA. These proportions were all higher than those in the control group(6.0%, 46.9%, and 2.9%, respectively; all P<0.001). In the high myopia group, ACD and ACD/AL were negatively correlated with AL(r=-0.162 and -0.661, respectively; all P<0.001), while both ACD and ACD/AL in the control group were positively correlated with AL(r=0.338 and 0.105, respectively; both P<0.01). In the high myopia group, CA increased with age when the patient's age was ≥50 years(r=0.197, P<0.001), which was consistent with the control group.CONCLUSION: The standardized ocular biometric data of cataract patients with high myopia before cataract surgery are helpful for ophthalmologists to accurately calculate the intraocular lens(IOLs)power and select the appropriate IOL type. The majority of high myopia patients need simultaneous correction of CA during cataract surgery.

ObjectiveIn order to improve the recognition accuracy of named entities in medical record texts and realize the effective mining and utilization of medical record knowledge， a Bert-Radical-Lexicon（BRL） neural network model is constructed to recognize medical record entities with respect to the characteristics of medical record texts. MethodWe selected 408 medical records related to hypertension from the the Complete Library of Famous Medical Records of Chinese Dynasties and constructed a dataset consisting of 1 672 medical records by manually labeling. Then， we randomly divided the dataset into three subsets， including the training set（1 004 cases）， the testing set （334 cases） and the validation set（334 cases）. Based on this dataset， we built a BRL model that fused various text features of medical records， as well as its variants BRL-B， BRL-L and BRL-R， and a baseline model Base for experiments. During the model training phase， we trained the above models using the training set to reduce the risk of overfitting. We continuously monitored the performance of each model on the validation set during training and saved the model with the best performance. Finally， we evaluated the performance of these models on the testing set. ResultCompared with other models， the BRL model had the best performance in the medical records named entity recognition task， with an overall recognition precision of 90.09%， a recall of 90.61%， and the harmonic mean of the precision and recall（F1） of 90.35% for eight types of entities， including disease， symptom， tongue manifestation， pulse condition， syndrome， method of treatment， prescription and traditional Chinese medicine（TCM）. Compared with the Base model， the BRL model improved the overall F1 value of entity recognition by 5.22%， and the F1 value of pulse condition entity increased by 6.92%， which was the largest increase. ConclusionBy incorporating a variety of medical record text features in the embedding layer， the BRL neural network model has stronger named entity recognition ability， and thus extracts more accurate and reliable TCM clinical information.

ObjectiveTo achieve high-dimensional prediction of class imbalanced of adverse drug reaction（ADR） of traditional Chinese medicine（TCM） and to classify and identify risk factors affecting the occurrence of ADR based on the post-marketing safety data of TCM monitored centrally in real world hospitals. MethodThe ensemble clustering resampling combined with regularized Group Lasso regression was used to perform high-dimensional balancing of ADR class-imbalanced data， and then to integrate the balanced datasets to achieve ADR prediction and the risk factor identification by category. ResultA practical example study of the proposed method on a monitoring data of TCM injection performed that the accuracy of the ADR prediction， the prediction sensitivity， the prediction specificity and the area under receiver operating characteristic curve（AUC） were all above 0.8 on the test set. Meanwhile， 40 risk factors affecting the occurrence of ADR were screened out from total 600 high-dimensional variables. And the effect of risk factors on the occurrence of ADR was identified by classification weighting. The important risk factors were classified as follows：past history， medication information， name of combined drugs， disease status， number of combined drugs and personal data. ConclusionIn the real world data of rare ADR with a large amount of clinical variables， this paper realized accurate ADR prediction on high-dimensional and class imbalanced condition， and classified and identified the key risk factors and their clinical significance of categories， so as to provide risk early warning for clinical rational drug use and combined drug use， as well as scientific basis for reevaluation of safety of post-marketing TCM.