To investigate the prognostic factors affecting survival in patients with natural killer T-cell lymphoma (NKTL), and then develop a prognostic model for predicting their overall survival (OS) based on random survival forest (RSF) algorithm.

Demographic and clinical pathological data of NKTL patients were collected from the SEER database during 2000 and 2020. The patients were divided into a training cohort (n=471) and a validation cohort (n=203) in a 7:3 ratio. Cox regression analysis was performed to identify prognostic factors affecting OS, and a nomogram model was constructed based on the obtained factors. Meanwhile, RSF algorithm was used to determine prognostic factors affecting OS to build the RSF model. The models were evaluated using receiver operating characteristic (ROC) curve, calibration curve, decision curve, net reclassification improvement (NRI), and integrated discrimination improvement (IDI), and the predictive performances of the 2 models were compared. Risk scores for each patient were calculated using the 2 models. Then the patients were divided into high- and low-risk groups based on the median risk score, and survival curve was plotted for comparison.

Ann Arbor stage, age, radiotherapy, combined treatment, and type of disease were identified as significant prognostic variables associated with OS. In the validation cohort, the area under the ROC curve (AUC) for the nomogram model at 1, 3, and 5 years was 0.745, 0.771, and 0.748, respectively, while the AUC for the RSF model was 0.764, 0.792, and 0.761 at the same time points. ROC curve analysis indicated that both models demonstrated good accuracy and discrimination in predicting OS. Calibration curve analysis showed a strong consistency between the predicted and actual OS for both models. Both models effectively stratified the patients into poor and favorable prognosis groups, with the OS of patients in the poor prognosis group being significantly shorter than that of the favorable prognosis group (P<0.0001). Decision curve analysis revealed that the net benefit of the RSF model was superior to that of the nomogram model. Compared to the nomogram model, the NRI for the RSF model was 0.184 (95%CI: 0.098~0.267, P<0.01), and the IDI was 0.300 (95%CI: 0.241~0.359, P<0.01). Overall, the RSF model demonstrated superior predictive capability than the nomogram model.

Ann Arbor stage, age, radiotherapy, combined treatment, and type of disease are prognostic factors affecting the prognosis of NKTL patients. Our RSF model demonstrates strong predictive capability for the prognosis of NKTL patients and can effectively assess patient outcomes.

To conduct an in-depth analysis and feature extraction of the transcriptomics data of 10 types of cancers in order to realize the staging diagnosis of cancer samples.

The transcriptomics data of the top 10 cancers having the highest incidence were amassed from the UCSC Xena website, which comprised 4938 samples and 59428 genes. With the aid of variational autoencoder, we developed an incremental feature ranking and selection variational autoencoder (IFRSVAE) based on feature importance ranking and incorporating the masking algorithm and the Incremental Feature Selection (IFS). Subsequently, the performance efficiency of our IFRSVAE model was evaluated in conjunction with Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGboost), and it was also compared with other methods.

Our research extracted 21 features for the ensuing classification. In comparison to the conventional variational autoencoder, recursive feature elimination, and Lasso regression models, the IFRSVAE model attained more favorable performance across all 3 classifiers (highest AUC value, and well performed other indicators). Notably, the IFRSVAE-RF exhibited the most outstanding performance, with an AUC value reaching 85.49% (95%CI: 83.24%~87.74%). Moreover, Shapley additive explanations (SHAP) interpretable model illustrated well contributions of the features in our model.

Our developed IFRSVAE shows certain effectiveness in feature extraction. The constructed IFRSVAE-RF model demonstrates relatively good performance in the task of cancer staging diagnosis, which providing a new and referable idea for research orientation of deep-learning-based diagnostic methods for cancer staging.

To perform in-depth integration and analysis of cancer multi-omics data and clinical data to enhance the predictive capability for cancer prognosis.

Multi-omics data (mRNA, lncRNA, and miRNA profiles) and clinical data for 3 cancer types, namely ovarian cancer, liver cancer, and colorectal cancer, were retrieved from The Cancer Genome Atlas (TCGA) database. A novel cancer prognosis prediction model, MoACG (Multi-omics Attention Clinical Gating Model), was constructed based on the self-attention mechanism to explore potential associations among different omics layers, and gated fusion was employed to adaptively integrate multi-omics information with clinical information(age, sex, and treatment modality). The effectiveness of the model was validated through comparison with multiple machine learning methods, and the interpretability algorithm DeepLIFT was utilized to quantify gene contributions to the model and identify core prognostic genes.

In the ovarian cancer, liver cancer, and colorectal cancer datasets, five-fold cross-validation yielded the area under curve (AUC) values of receiver operating characteristic (ROC) curve of (0.793±0.042), (0.791±0.065), and (0.789±0.086), respectively, and the AUC values of precision-recall curve (AUPR) were (0.915±0.020), (0.855±0.058), and (0.917±0.039), respectively. The comprehensive performance surpassed that of 9 other machine learning models. Ablation experiments demonstrated that the 3-omics data integration model exhibited optimal predictive performance across all cancer types. The DeepLIFT algorithm identified MED8, DLGAP4, and NABP2 as genes associated with liver cancer, showing high concordance with existing research findings and effectively stratifying patient survival risk based on expression levels (P<0.005).

Compared with previous studies, the MoACG model, constructed by integrating multi-omics and clinical data, effectively enhances the predictive performance for cancer prognosis, thereby providing a novel approach for cancer diagnosis, treatment, and prognostic research.

To construct a time series analysis fusion tool using multisource internet data and then accurately predict the incidence trend of hepatitis in Chongqing.

The incidence rate of hepatitis were obtained from the database of the Centre for Health and Disease Control. Air pollutant data were obtained from the official website of the China Environmental Monitoring Station, climate data were obtained from the National Meteorological Galaxy Center, and network index data were obtained through Baidu search engine. The time duration was from November 2013 to May 2023. Based on existing time series analysis methods, multisource data were used to correct the residual part of the decomposition model. A delayed input neural network (DINN) was constructed based on the respective advantages of non autoregressive (NAR) and long short-term memory (LSTM) recurrent neural networks. Afterwards, optimization modules such as the Nutcracker Optimization Algorithm (NOA) and Joint Quantile Huber Loss (JQHL) were added to the foundation, and then DINN + was constructed.

Compared to common single-input models and synchronous multi-input models, DINN achieved the best prediction performance. After adding hyperparameters and loss function optimization, the predictive performance of DINN+ was further improved, with a mean-square error (MSE) of 0.170 9, a mean absolute error (MAE) of 0.461 2, a root-mean-square error (RMSE) of 0.582 1, a mean absolute percentage error (MAPE) of 0.062 6, and a R-square (R²) of 0.884 0 in a testing set.

Based on the ideas of diverse methods and multidimensional data fusion, we propose a DINN+ optimization model with good accuracy and generalization ability on the basis of previous time series analysis. This model enriches and supplements the methodological research content of using multisource data to calibrate infectious disease time series prediction analysis and can serve as a new benchmark for future analysis of influencing factors and trend prediction of infectious disease public health.

To develop a deep learning-based multi-dimensional segmentation system to recognize small lesions in magnetic resonance imaging（MRI）images in order to provide a decision-making basis for the diagnosis and treatment of acute ischemic stroke（AIS）.

We extracted and fused the features from 2D and 3D network, introduced the joint loss function, and then proposed a new 2.5D method, a multi-dimensional multi-scale attention enhanced network（MMAE-Net）. On AIS segmentation datasets（171 cases in training set and 43 cases in testing set）, the proposed method was trained and tested, and its performance was compared with other methods.

When compared to 2D and 3D networks, our 2.5D network（MMAE-Net）achieved the best segmentation performance in all evaluation indicators, with a dice similarity coefficient（DSC）of 81.25% and a sensitivity of 84.82%. MMAE-Net achieved better segmentation performance when compared to U-Net, ResU-Net, DenseU-Net, AttentionU-Net, Segmentation TRansformer（SETR）, and other classical methods and previous research. In addition, we also created a visual and automated clinical application system to improve the practical and promotive capability of methods.

Based on fusing the features from 2D and 3D network, a 2.5D multi-dimensional segmentation model MMAE-Net is developed, which has achieved excellent performance in the recognition of MRI small lesions and provides an effective solution for the diagnosis and treatment of AIS diseases.

To build a stacked supervised autoencoder（SSAE）model based on transcriptomic data, so as to improve the prognostic prediction of lung adenocarcinoma（LUAD）.

Transcriptomic data（475 samples and 25 481 genes）from the Cancer Genome Atlas（TCGA）database were collected, and the survival prognosis and gene differential expression analyses were performed in LUAD patients, using SSAE, random survival forest（RSF）, and DeepSurv methods, respectively. Concordance index（CI）and P-value of Log-rank test were adopted to evaluate the performance of each method.

SSAE had a higher concordance index（CI=0.58）and a lower Log-rank test P value（P=0.05）than RSF（CI=0.54, P=0.15）and DeepSurv（CI=0.55, P=0.10）. There were significant differences in survival outcomes between the high-risk and low-risk groups in the survival analysis（HR=2.841; 95%CI: 1.907~4.232; Log-rank test P<0.001）. Biogenic analysis identified 21 representative differentially upregulated genes, including IGFBP1, ANXA13, MUC2, CIDEC, NTSR1 and DSG3.

SSAE with omics data significantly improves the prognostic prediction of LUAD. The cross-fusion of deep learning and omics research provides a novel scheme for cancer-related research of diagnosis, treatment, and prognosis.

To compare and analyze clinical trial design protocols and results of 6 clinical trials concerning revascularization in acute stroke due to anterior circulation large vessel occlusion in order to improve clinical researchers' comprehensive understanding on the importance of clinical trial design.

As for the key points of clinical trial design, rationality and scientificity of the designs in term of study population, control selection, main indexes, comparison type and sample size estimation were analyzed.

There were some problems in the design of 6 clinical trials. As a result, the main indexes of the 4 trials did not meet the significance test standards. The main problems were as following: study population inconsistent with the study objectives, adoption of unreasonable superiority comparison, unreasonable selection for parameters of sample size estimation, and non-inferiority margin setting not meeting the minimum clinically important difference.

There are still many misunderstandings in clinical trial design, which need further awareness and attentions.