AiJun Peng

Department of Neurosurgery, Affiliated Hospital of Yangzhou University, Yangzhou University, No.368, Hanjiang Middle Road, Hanjiang district, Yangzhou city, Jiangsu province, 225000, China. 13801456336@163.com.

Dr. AiJun Peng is a neurosurgeon affiliated with the Department of Neurosurgery at West China Hospital, Sichuan University, located in Chengdu, Sichuan, China. His research interests include the application of radiomics and machine learning in neuroimaging, particularly in the differentiation of sellar region lesions. Dr. Peng has co-authored studies focusing on advanced imaging analysis techniques to improve diagnostic accuracy in neurosurgical practice.

For more detailed information about his publications and research contributions, you can visit his ORCID profile: https://orcid.org/0000-0002-8700-7736.

• Radiomic study of common sellar region lesions differentiation in magnetic resonance imaging based on multi-classification machine learning model
• Valproic acid promotes transcriptional activation of Drd2 by mediating histone acetylation to inhibit the mTOR-Pttg1 signaling axis and exerts anti-PitNETs activity
• A semantic segmentation model for automatic precise identification of pituitary microadenomas with preoperative MRI
• Exploring the associations of gut microbiota with inflammatory and the early hematoma expansion in intracerebral hemorrhage: from change to potential therapeutic objectives
• Gut Microbiome variation in patients with early-stage mild-to-moderate intracerebral hemorrhage: A pilot study exploring therapeutic targets
• Dysbiosis of the gut microbiota in glioblastoma patients and potential biomarkers for risk assessment
• Integrating Clinical Data and Radiomics and Deep Learning Features for End-to-End Delayed Cerebral Ischemia Prediction on Noncontrast CT
• Naringenin Alleviates Radiation-Induced Intestinal Injury by Inhibiting TRPV6 in Mice

Hang Qu et al and the neurosurgeons LiangXue Zhou from the West China Hospital, and AiJun Peng from the Affiliated Hospital of Yangzhou University in a Retrospective radiomics-based diagnostic performance study using supervised machine learning studied two hundred and fifty-eight pathologically diagnosed sellar region lesions, including 54 TSMs, 81 CRs, 61 RCCs, and 63 PAs. All patients underwent conventional MR examinations. Feature extraction and data normalization, and balance were performed. Extreme gradient boosting (XGBoost), support vector machine (SVM), and logistic regression (LR) models were trained with the radiomics features. Five-fold cross-validation was used to evaluate model performance.

The XGBoost model showed better performance than the SVM and LR models built from contrast-enhanced T1-weighted MRI features (balanced accuracy 0.83, 0.77, 0.75; AUC 0.956, 0.938, 0.929, respectively). Additionally, these models demonstrated significant differences in sensitivity (P = 0.032) and specificity (P = 0.045). The performance of the XGBoost model was superior to that of the SVM and LR models in differentiating sellar region lesions by using contrast-enhanced T1-weighted MRI features.

The proposed model has the potential to improve the diagnostic accuracy in differentiating sellar region lesions ¹⁾.

• Innovative Use of Radiomics: The application of quantitative radiomic features to differentiate complex sellar pathologies is timely and clinically relevant, especially given the overlapping imaging characteristics of these lesions.

• Multi-class Design: Unlike many prior works focused on binary classification (e.g., tumor vs. no tumor), this study addresses a more challenging and realistic multi-classification problem (TSM vs. CR vs. RCC vs. PA).

• Robust Validation: The use of five-fold cross-validation enhances the reliability of the reported results and guards against overfitting, especially important in retrospective studies.

• Comparative Modeling: The inclusion of three distinct supervised learning methods allows a meaningful comparison, where XGBoost showed superior balanced accuracy (0.83) and AUC (0.956), suggesting strong discriminative capability using only contrast-enhanced T1-weighted MR images.

• Limited Imaging Modalities: The study relies solely on contrast-enhanced T1-weighted MRI. Including additional sequences (e.g., T2, FLAIR, DWI) might have enriched the radiomics dataset and improved performance or generalizability.

• Single-Center Retrospective Design: The study’s retrospective and likely single-institution nature may introduce selection bias and limit external validity. Prospective multicenter validation is needed before clinical deployment.

• Unclear Feature Interpretability: While XGBoost outperformed other models, the interpretability of the selected radiomic features is not deeply discussed. This limits the translational impact for clinicians who need to understand why the model makes certain predictions.

• No External Test Set: Despite cross-validation, the absence of an external validation cohort means model generalizability remains unproven in real-world unseen data.

The study supports the utility of advanced machine learning techniques, especially Extreme gradient boosting, for non-invasive differentiation of sellar region lesions. If externally validated, such models could aid radiologists and neurosurgeons in prioritizing differential diagnoses, potentially reducing diagnostic delays or unnecessary interventions.

This study represents a solid contribution to the growing field of radiomics and machine learning in neuroimaging. The use of multi-classification modeling and comparative performance evaluation is commendable. Future directions should include multi-sequence MRI integration, prospective validation, and improved model explainability.