Cancer driver genes play a key role in tumor development, and the mutation or aberrant expression of these genes, which mainly occurs at the somatic cell level, has become a focus of research in recent years. Identifying lung cancer driver genes from millions of somatic cell mutations is undoubtedly a challenging task. In order to accurately predict lung cancer driver genes, researchers have proposed numerous computational methods, most of which employ machine learning techniques such as random forests and support vector machines (SVMs). However, simply merging features extracted from different data sources and using the selected features for training machine learning models is not an optimal strategy. New research points out that gene expression data can effectively characterize the similarity between genes. Based on this finding, an advanced deep learning-based analysis method, BiLSTM-Driver is proposed which significantly improves the prediction accuracy of lung cancer driver genes by probing deeply into the mutational features of genes and their neighbors through the Bidirectional Long Short-Term Memory networks. This method performs well in lung cancer research, especially in the prediction of lung adenocarcinoma and lung squamous carcinoma, and achieves better performance in AUC performance, precision and recall, achieving the performance of 0.994,0.991, and 0.987, and 0.987,0.993, and 0.991,respectively, which is significantly better than other algorithms. The proposed method BiLSTM-Driver provides an innovative way to identify new lung cancer driver genes. Through in-depth analysis of lung cancer-related genomic data, BiLSTM-Driver is able to accurately identify potential oncogenes and reveal their key roles in lung cancer development.

Non-Small Cell Lung Cancer(NSCLC) is characterized by poor prognosis and high mortality rates, making the prediction of survival duration crucial for NSCLC patients. Traditional cancer survival prediction relies mainly on the analysis of clinical and pathological characteristics, but these pieces of information often fail to fully reveal the complexity of the disease, thereby limiting the accuracy and reliability of survival prediction. Current deep learning techniques face several challenges, including inadequate utilization of data modalities, only predicting patient survival outcomes, and predicting shorter survival periods for patients. To address these issues, a multimodal fusion neural network, MTPSN (Multimodal Transformer and Convolutional Neural Network for Survival Prediction in Non-Small Cell Lung Cancer Patients), was designed. This model combines convolutional neural networks (CNNs) and transformers to predict the survival duration of NSCLC patients. The model uses CNNs to process multi-modal and high-dimensional genetic data, including clinical information, mRNA, miRNA, copy number variations, and DNA methylation, effectively extracting feature representations. By integrating and fusing these multi-modal data features, the model can utilize the fused feature representations to predict the survival duration of NSCLC patients. Experimental results demonstrate that on squamous cell carcinoma and adenocarcinoma datasets, the proposed survival prediction method achieves a Concordance Index (C-index) of 0.748 and an Integrated Brier Score (IBS) of 0.172, outperforming existing survival prediction models significantly.

In recent years, the incidence of stroke has significantly increased, posing a serious threat to public health. Accurate stroke lesion segmentation techniques can assist physicians in promptly formulating appropriate treatment plans based on specific patient conditions, significantly reducing the risk of disability and mortality, thereby improving patient outcomes. Against this backdrop, leveraging its powerful representation and reasoning capabilities, deep learning has emerged as a key research direction in the field of medical image processing. However, deep learning-based stroke lesion segmentation methods rely on a large amount of precisely labeled medical imaging data, the acquisition of which often faces challenges such as high costs, insufficient quantities, and time-intensive efforts. Traditional data augmentation methods provide some relief but are still limited by data distribution and diversity constraints. Addressing these issues, this paper introduces a novel data generation model, Stroke-CVAE-CGAN, which generates “synthetic lesion masks” based on the spatial distribution characteristics of stroke lesions, serving as constraints for Conditional Generative Adversarial Networks (CGANs), thus enabling the augmented training data that aligns with the distribution patterns of real stroke lesions. Experiments conducted on the ATLAS stroke lesion segmentation dataset show that the augmented data generated by Stroke-CVAE-CGAN closely matches the training data in terms of distribution and exhibits superior Frechet Inception Distance (FID) quality. Utilizing this augmented data to train the U-Net segmentation model significantly enhances the accuracy of stroke lesion segmentation.