Machine Learning: A Novel Approach for Predicting Visual Outcomes and Factors Affecting it in Patients with Pituitary...

To use machine learning tool to predict visual outcomes.

A retrospective cohort of 284 consecutive pituitary adenoma patients with preoperative visual deficit was used. Patient variables were collected. Preprocessing and classification was done in the open source ML tool box Weka (Ver 3.8.4). Four algorithms were used for classification-the J48 trees, LMT algorithm, the REP tree algorithm, and the Random Forest.

The study included a total of 284 patients. The mean duration of visual symptoms was 10.09 ± 4.3 months. There were 24 patients with uniocular involvement and 260 patients with binocular visual impairment: Vision improved in 89.78% patients, remained same in 9.86% and deteriorated in 0.3% patients and were same in 9.86% patients and deteriorated in 0.3% patient. Factors like extent of resection, preoperative visual acuity, tumor volume, and duration of symptoms were found to have statistically significant effect on postoperative visual outcome. The model predicting visual improvement had an AUC of 0.846; 88.94% accuracy, 78.4% specificity, and 93.7% sensitivity; 90.5% and 85.04% positive and negative predictive value, respectively; an F1 score of 0.921; and a Brier score of 0.291.

Postoperative visual outcomes of pituitary adenoma surgery and factors affecting it can be predicted with 88.94% accuracy using a machine learning approach. Based on this preliminary evaluation, ML appears promising in predicting visual outcomes after endoscopic resection of pituitary adenomas, thus improving patient-tailored care and counselling.