Min Chen,Yuxin Li,Sumei Zhou,Linbo Zou,Lei Yu,Tianfang Deng,Xian Rong,Shirong Shao,Jijun Wu
Abstract:To construct a prediction model of olfactory dysfunction after transnasal sellar pituitary tumor resection based on machine learning algorithms. A cross-sectional study was conducted. From January to December 2022, 158 patients underwent transnasal sellar pituitary tumor resection in three tertiary hospitals in Sichuan Province were selected as the research objects. The olfactory status was evaluated one week after surgery. They were randomly divided into a training set and a test set according to the ratio of 8:2. The training set was used to construct the prediction model, and the test set was used to evaluate the effect of the model. Based on different machine learning algorithms, BP neural network, logistic regression, decision tree, support vector machine, random forest, LightGBM, XGBoost, and AdaBoost were established to construct olfactory dysfunction risk prediction models. The accuracy, precision, recall, F1 score, and area under the ROC curve (AUC) were used to evaluate the model's prediction performance, the optimal prediction model algorithm was selected, and the model was verified in the test set of patients. Of the 158 patients, 116 (73.42%) had postoperative olfactory dysfunction. After missing value processing and feature screening, an essential order of influencing factors of olfactory dysfunction was obtained. Among them, the duration of operation, gender, type of pituitary tumor, pituitary tumor apoplexy, nasal adhesion, age, cerebrospinal fluid leakage, blood scar formation, and smoking history became the risk factors of olfactory dysfunction, which were the key indicators of the construction of the model. Among them, the random forest model had the highest AUC of 0.846, and the accuracy, precision, recall, and F1 score were 0.750, 0.870, 0.947, and 0.833, respectively. Compared with the BP neural network, logistic regression, decision tree, support vector machine, LightGBM, XGBoost, and AdaBoost, the random forest model has more advantages in predicting olfactory dysfunction in patients after transnasal sellar pituitary tumor resection, which is helpful for early identification and intervention of high-risk clinical population, and has good clinical application prospects.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to establish a prediction model through machine - learning algorithms for predicting the risk of olfactory disorders in patients after transsphenoidal pituitary tumor resection. Specifically, the research aims to use a variety of machine - learning algorithms (such as BP neural network, logistic regression, decision tree, support vector machine, random forest, LightGBM, XGBoost and AdaBoost) to construct a risk prediction model for olfactory dysfunction, and by comparing the performance of different models, select the optimal prediction model. This helps in the early identification of high - risk groups, so as to take intervention measures and improve the clinical treatment effect and quality of life of patients.
### Research Background
Pituitary tumors are a common benign tumor, accounting for about 10 - 20% of intracranial tumors, second only to meningiomas. With the development of medical technology, especially the progress of optical and electrophysiological techniques, transsphenoidal endoscopic pituitary tumor resection has become the preferred method for treating pituitary tumors. Although this type of surgery has less trauma and faster recovery, still 10.5% - 44% of patients have olfactory disorders after surgery, which is due to the inevitable damage to the nasal mucosa, nasal septum and inferior turbinate during the surgical process.
### Research Purposes
1. **Construct a prediction model**: Based on machine - learning algorithms, construct a risk prediction model for olfactory disorders after transsphenoidal pituitary tumor resection.
2. **Evaluate model performance**: Evaluate the prediction performance of different models through indicators such as accuracy, precision, recall, F1 - value and the area under the ROC curve (AUC).
3. **Select the optimal model**: By comparing the performance of different models, select the optimal prediction model to improve the prediction accuracy of olfactory disorders.
### Main Findings
- **Data set**: A total of 158 patients who underwent transsphenoidal pituitary tumor resection were included in the study, among which 116 cases (73.42%) had olfactory disorders after surgery, and 42 cases (26.58%) did not have olfactory disorders.
- **Influencing factors**: Operation time, gender, pituitary tumor type, pituitary tumor apoplexy, nasal cavity adhesion, cerebrospinal fluid leakage, blood scab formation, age and smoking history were determined as the main risk factors for olfactory disorders.
- **Optimal model**: The random forest model performed the best, with an AUC value of 0.846, an accuracy of 0.750, a precision of 0.870, a recall of 0.947, and an F1 - value of 0.833.
### Conclusion
The prediction model constructed through machine - learning algorithms, especially the random forest model, can effectively predict the risk of olfactory disorders in patients after transsphenoidal pituitary tumor resection, which helps in the early identification of high - risk patients, provides more effective clinical decision - making support, and thus improves the quality of life and overall recovery of patients.