Endocrine Abstracts

Background: Existing biomarkers have limited ability to discriminate indolent non-functioning pituitary adenomas (NFPAs) from those with a propensity to recur following primary surgery. Radiomics, the extraction of quantitative data from medical imaging, is increasingly recognised as a tool to augment clinical decision making.

Methods: 39 patients who underwent primary trans-sphenoidal surgery for an NFPA between January 2007 and April 2017, were enrolled. 19 patients required multiple therapeutic interventions (revision surgery and/or radiotherapy) or demonstrated radiological recurrence / regrowth following TSS. 20 patients demonstrated indolent behaviour, with stable post-operative appearances at a minimum of 3 years following surgery [36-111 months follow-up]. The tumour was manually segmented on pre-operative T1SE MR images using 3D Slicer. 135 radiomic features were extracted from the images using Python module pyradiomics via 3D Slicer module ‘Radiomics’.

Results: Neighbouring Gray Tone Difference Matrix (NGTDM) Coarseness, a measure of the spatial rate of change of pixel intensity, was the strongest predictive feature. Receiver operating characteristic curve (ROC) analysis showed an area under the curve of 0.767 [0.603 – 0.887]. Kaplan-Meier analysis, using the optimal criterion from the ROC, showed a difference in mean progression free survival of 105.7 vs 43.4 months (P<0.0001). Logistic regression analysis was performed to ascertain the combined predictive effect of 1) Coarseness, 2) Inverse Difference Normalised and 3) Maximum 2D diameter. The logistic regression model was statistically significant: χ2 17.88, P < 0.001 The model correctly identified 74.4% of cases, with a positive predictive value of 80% for identifying more indolent cases.

Conclusions: In this pilot study we have shown that radiomic analyses have the potential to predict which NFPAs are likely to recur at an earlier stage, thereby potentially informing decision making in a context where biomarkers are lacking. Larger studies, with standardised image acquisition and processing are required to validate these findings.