Machine Learning-Based Radiomics of the Optic Chiasm Predict Visual Outcome Following Pituitary Adenoma Surgery

Yang Zhang, Chaoyue Chen, Wei Huang, Yangfan Cheng, Yuen Teng, Lei Zhang, Jianguo XuJournal of Personalized Medicine 2021 · DOI

TL;DR

This paper explores using machine learning and radiomics on pre-operative MRI scans of the optic chiasm to predict visual recovery after pituitary adenoma surgery. Their SVM-based model achieved an AUC of 0.824, suggesting a promising new tool for personalized patient counseling.

What Problem Does This Paper Address?

Pituitary adenomas, tumors in the pituitary gland, often compress the optic chiasm, leading to visual field defects like bitemporal hemianopia. While surgery can decompress the chiasm and improve vision, predicting the extent of visual recovery post-surgery is challenging. Current predictors are inconsistent, leaving a clinical need for a reliable method to identify patients who will regain vision and those who may experience permanent damage. This gap hinders personalized patient counseling and management.

Methods

The researchers employed a machine learning pipeline to analyze radiomic features extracted from pre-operative MRI scans of the optic chiasm.

  • Radiomic Feature Extraction: They used the PyRadiomics package to extract a comprehensive set of 851 features from coronal T2-weighted MRI images of the optic chiasm. These features capture quantitative information about the shape, intensity distribution, and texture of the optic chiasm.
  • Feature Selection: To handle the high dimensionality of radiomic features and reduce redundancy, they utilized Least Absolute Shrinkage and Selection Operator (LASSO) regression.
  • Model Development: Three machine learning algorithms were trained to predict visual recovery: Support Vector Machine (SVM), Random Forest (RF), and Linear Discriminant Analysis (LDA).
  • Model Evaluation: The performance of these models was assessed using five-fold cross-validation, a robust technique for evaluating model generalization. Key metrics like Area Under the Curve (AUC), accuracy, sensitivity, and specificity were calculated.

Datasets

The study utilized a retrospective dataset of 131 pituitary adenoma patients from the West China Hospital, Sichuan University. These patients underwent surgical chiasmal decompression between January 2017 and June 2019. The dataset was divided into two groups based on post-operative visual field outcome: a "recovery group" (79 patients) and a "non-recovery group" (52 patients).

Key Insights and Ideas

  • Radiomics as a Biomarker for Axonal Injury: The study suggests that radiomic features of the optic chiasm can reflect underlying microstructural changes in axons, such as ischemic injury and demyelination, which are crucial for visual impairment in these cases. This extends the application of radiomics beyond tumor characterization to predicting functional outcomes.
  • SVM's Superiority in this Context: The SVM model outperformed RF and LDA, potentially due to its robustness against overfitting and its ability to model non-linear relationships between complex radiomic features and visual recovery. This highlights the importance of algorithm selection in radiomic studies.
  • Clinical Utility of Routine Imaging: The radiomic features were extracted from standard T2-weighted MRI sequences, which are routinely acquired during pre-operative evaluations. This implies that the proposed predictive model could be integrated into clinical practice without requiring additional imaging protocols or costs.

Key Results

  • All three machine learning models (SVM, RF, LDA) demonstrated feasible performance in predicting visual field recovery, achieving an AUC greater than 0.750.
  • The SVM-based radiomic model achieved the highest predictive performance, with an AUC of 0.824 in five-fold cross-validation.
  • The LDA and RF models showed AUCs of 0.801 and 0.751, respectively.
  • The study found no significant differences in age, sex distribution, or surgical approach between the recovery and non-recovery groups, underscoring the need for more sophisticated predictive tools.

Limitations and Future Work

The authors acknowledge several limitations:

  • Retrospective and Single-Center Design: This introduces potential selection bias. Future prospective, multi-center studies are recommended to validate these findings.
  • Variable Follow-up Periods: While the impact is considered minimal, visual outcome was not measured at a uniform time point post-surgery.
  • Invisibility of Optic Chiasm: The model cannot be applied to cases where the optic chiasm is obscured by severe tumor compression on pre-operative MRI.
  • Missing OCT Data: Optical Coherence Tomography (OCT) data, specifically retinal nerve fiber layer (RNFL) thickness, which is a known factor in visual recovery, was not routinely collected and thus not incorporated. Future studies could explore its inclusion.
  • Classical ML Algorithms: The study used established machine learning algorithms. Future research could explore more advanced or novel algorithms to further improve predictive accuracy.

Despite these limitations, the study concludes that machine learning-based radiomics of the optic chiasm holds significant potential as a non-invasive, pre-operative tool for predicting visual recovery after pituitary adenoma surgery, paving the way for more personalized patient care.

Summary updated

Abstract

Preoperative prediction of visual recovery after pituitary adenoma surgery remains a challenge. We aimed to investigate the value of MRI-based radiomics of the optic chiasm in predicting postoperative visual field outcome using machine learning technology. A total of 131 pituitary adenoma patients were retrospectively enrolled and divided into the recovery group (N = 79) and the non-recovery group (N = 52) according to visual field outcome following surgical chiasmal decompression. Radiomic features were extracted from the optic chiasm on preoperative coronal T2-weighted imaging. Least absolute shrinkage and selection operator regression were first used to select optimal features. Then, three machine learning algorithms were employed to develop radiomic models to predict visual recovery, including support vector machine (SVM), random forest and linear discriminant analysis. The prognostic performances of models were evaluated via five-fold cross-validation. The results showed that radiomic models using different machine learning algorithms all achieved area under the curve (AUC) over 0.750. The SVM-based model represented the best predictive performance for visual field recovery, with the highest AUC of 0.824. In conclusion, machine learning-based radiomics of the optic chiasm on routine MR imaging could potentially serve as a novel approach to preoperatively predict visual recovery and allow personalized counseling for individual pituitary adenoma patients.
BibTeX
@article{zhang2021machine,
  author = {Yang Zhang and Chaoyue Chen and Wei Huang and Yangfan Cheng and Yuen Teng and Lei Zhang and Jianguo Xu},
  title = {{Machine Learning-Based Radiomics of the Optic Chiasm Predict Visual Outcome Following Pituitary Adenoma Surgery}},
  journal = {Journal of Personalized Medicine},
  year = {2021},
  doi = {10.3390/jpm11100991},
}
Figure