Machine Learning Modeling the Risk of Refractory Pituitary Adenoma Using Radiomic and Pathomic Data

Refractory pituitary adenoma is characterized by invasive tumor growth, continuous growth and/or hormone hypersecretion in spite of standardized multi-modal treatment such as surgeries, medications or radiations. Quality of life or even lives are threatened by these tumors. According to the 2017 World Health Organization's new classification guideline of pituitary adenoma, patients have to suffer from symptoms or complications caused by these tumors, to bear a heavy financial burden, and to accept additional therapeutic side effects when the diagnosis of "refractory pituitary adenoma" is made. If refractory pituitary adenoma could be predicted at early stage, these patients would be able to have a more frequent clinical follow-up, receive multiple effective treatment as early as possible, or even be enrolled in clinical trials of investigational medications, so as to prevent or delay the recurrence or persistent of the tumor growth. Therefore, the unmet clinical need falls into an early prediction system for refractory pituitary adenomas, which could provide accurate guidance for subsequent treatment in the early stage. The investigators have constructed a pituitary adenoma database including clinical data, radiological images, pathological images and genetic information. The investigators are proposing a study using machine learning to extract features from these multi-dimensional, multi-omics data, which could be further used to train a prediction model for the risk of refractory pituitary adenoma. The proposed model would also be validated in another prospectively collected database. The established model would be able to identify potential medication targets and provide guidance for personalized therapy of refractory pituitary adenoma.