An artificial intelligence system for estimating the improvement of clinical and paraclinical parameters after therapy in pituitary tumors

Ana-Silvia Corlan; Babuc Diogen; Costi Flavia; Mihaela Vlad; Melania Balas; Ioana Golu; Daniela-Georgiana Amzar; Onchis Darian

doi:10.1530/endoabs.99.EP695

EP695

Prev Next

Section Contents Cite

Endocrine Abstracts (2024) 99 EP695 | DOI: 10.1530/endoabs.99.EP695

ECE2024 Eposter Presentations Pituitary and Neuroendocrinology (214 abstracts)

An artificial intelligence system for estimating the improvement of clinical and paraclinical parameters after therapy in pituitary tumors

Ana-Silvia Corlan ¹ , Babuc Diogen ² , Costi Flavia ² , Mihaela Vlad ¹ , Melania Balas ¹ , Ioana Golu ¹ , Daniela-Georgiana Amzar ¹ & Onchis Darian ²

Views

Author affiliations

¹University of Medicine and Pharmacy Timisoara, Endocrinology, Timisoara, Romania;²West University of Timisoara, Computer Science, Timisoara, Romania

Pituitary tumors usually have an excellent prognosis after targeted treatment involving surgery, radiotherapy and medical therapy. Metabolic abnormalities are encountered in 22.3-52.5% of pituitary adenoma patients and are correlated with disease progression and prognosis. Therefore, in this study, we propose an artificial intelligence system for estimating the improvement of clinical and paraclinical parameters for patients undergoing therapy. The patient cohort is made out of 107 patients with a panel of clinical and paraclinical parameters recorded before (for 45 patients) and after therapy (for 62 patients). We first made a classification of tumor activity in case of functioning pituitary tumors. Secondly, we made a classification, distinguishing between symptomatic and asymptomatic patients for non-functioning pituitary adenomas. We trained a supervised deep neural network having as input the variable parameters of remaining tumor size after therapy, total volume and the clinical (systolic, diastolic blood pressure and body mass index) and paraclinical parameters (fasting plasma glucose, total cholesterol, LDL-cholesterol, HDL-cholesterol, triglycerides, uric acid, aspartate aminotransferase, alanine aminotransferase and C-reactive protein serum levels), and the fixed parameter treatment. Our target was the estimation of a cut-off value for symptomatic pituitary tumors. The versatile artificial intelligence system reached an accuracy of 85.71% for the patients after treatment, giving to the patient and the attending physician, valuable information regarding the prospective success of the treatment.