AI 临床医学课题【总结】

最近参与了几个临床医学课题，总结一下如何跨界结合

1： 确定研究的方向： 这个是决定文章的核心

       研究方向的时候，就要确定要投的期刊，平时看论文的时候要把一些常用的术语记录下来，

       投的期刊，研究内容，方法记录一下。

2：  研究团队团队搭建（负责人：负责读论文,研究点 ,确定方案 

                                        程序员：负责代码实现）

3：  定期的项目推进,复盘

如下是遇到的问题，Research Gate 讨论交流过程

Study Summary 
This study aims to develop a machine learning model for personalized prediction of overall survival (OS) in lymphoma patients using retrospective data. The model incorporates input features such as survival time, treatment regimens, demographic characteristics, and laboratory test results to predict a binary outcome (alive vs. deceased). Once trained, the model is intended for clinical use, where patient-specific features (including dynamically adjusted survival time) are input to generate real-time survival probability estimates.  
Key Methodological Questions 
1. Modeling Approach 
   - Is using survival/death as a binary outcome while also including survival time as an input feature the optimal strategy?  
   - How should censored patients (e.g., lost to follow-up) be handled in this framework?  
2. Treatment-Related Features 
   - Can treatment-related variables (e.g., specific regimens) be legitimately included as predictive features for survival outcomes?  
   - Does this introduce confounding due to treatment selection bias?  
3. Overfitting Concerns in Small Sample Size  
   - With only 70 samples, an internal test set, and no external validation, how can we rigorously assess and mitigate overfitting?  
   - What strategies (e.g., feature selection, regularization, or alternative validation methods) would be most effective?  

Answer: 

Shafagat Mahmudova

• PhD degree in Technical sciences, associate Professor
• Head of Department at Institute of Information Technology

The development of cancer is a complex process that occurs when genetic and epigenetic changes accumulate in the deoxyribose nucleic acid (DNA) of a cell. This leads to uncontrolled cell growth and invasion, which can ultimately result in the formation of a tumor. To better understand this disease and improve patient outcomes, researchers have traditionally relied on statistical and computational methods to analyse large datasets containing genomic, proteomic, and clinical information. However, with the emergence of artificial intelligence (AI) and ML, scientists are now able to develop