The pituitary adenoma (PA) resection strategy necessitates preoperative sellar area monitoring. Radiomics forecasting necessitates high-quality segmentation. Manual demarcation takes time and is prone to rater variability. The goal of this research is to develop an automated segmentation approach for the sellar area, as well as numerous tools for extracting invasiveness-related characteristics and evaluating their clinical value by predicting tumour consistency. At Peking Union Medical College Hospital, patients were diagnosed with pituitary adenoma. To automatically partition the sellar area into eight classes, a deep convolutional neural network dubbed gated-shaped U-net (GSU-Net) was developed. The segmentation findings were used to extract five MRI features: tumour diameters, volume, optic chiasma height, Knosp grading system, and degree of internal carotid artery contact. The diagnostic accuracy of the tumour consistency was used to assess the clinical utility of the offered techniques. The first group consisted of 163 individuals with verified pituitary adenoma who were randomly assigned to a training data set and a test data set. The second group consisted of fifty individuals with verified acromegaly. For the prediction of five invasive-related MRI characteristics, the suggested approaches obtained accuracies of greater than 80%. Approaches developed from automated segmentation outperformed original methods, with areas under the curve of 0.840 and 0.920 for clinical and radiomics models, respectively.
The suggested techniques could segment the sellar area automatically and extract characteristics with high accuracy. The excellent performance of the tumour consistency prediction suggests the techniques’ clinical use in assisting neurosurgeons in assessing patients’ situations, forecasting prognosis, and other downstream duties during the preoperative period.