Radiomics using multiparametric magnetic resonance imaging to predict postoperative visual outcomes of patients with pituitary adenoma.

BACKGROUND

Preoperative prediction of visual outcomes following pituitary adenoma surgery is challenging yet crucial for clinical decision-making. We aimed to develop models using radiomics from multiparametric MRI to predict postoperative visual outcomes.

METHODS

A cohort of 152 patients with pituitary adenoma was retrospectively enrolled and divided into recovery and non-recovery groups based on visual examinations performed six months after surgery. Radiomic features of the optic chiasm were extracted from preoperative T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and contrast-enhanced T1-weighted imaging (T1CE). Predictive models were constructed using the least absolute shrinkage and selection operator wrapped with a support vector machine through five-fold cross-validation in the development cohort and evaluated in an independent test cohort. Model performance was evaluated using the area under the curve (AUC), accuracy, sensitivity, and specificity.

RESULTS

Four models were established based on radiomic features selected from individual or combined sequences. The AUC values of the models based on T1WI, T2WI and T1CE were 0.784, 0.724, 0.822 in the development cohort, and 0.767, 0.763, 0.794 in the independent test cohort. The multiparametric model demonstrated superior performance among the four models, with AUC of 0.851, accuracy of 0.832. sensitivity of 0.700, specificity of 0.910 in the development cohort, and AUC of 0.847, accuracy of 0.800, sensitivity of 0.882 and specificity of 0.750 in the independent test cohort.

CONCLUSION

The multiparametric model utilizing radiomics of optic chiasm outperformed single-sequence models in predicting postoperative visual recovery in patients with pituitary adenoma, serving as a novel approach for enhancing personalized treatment strategies.