Radiomics of pituitary adenoma using computer vision: a review.

Abstract: Objective — to determine the role of preoperative neuroimaging parameters of invasiveness and consistency of pituitary adenoma. Material and methods. The present clinical study included 100 patients with a histologically confirmed diagnosis of pituitary adenoma. The average age of the subjects was 54.7 ± 18.8 years. At the preoperative stage, the diagnosis was based on clinical data, laboratory results, and neuroimaging research methods. The present study included an analysis of MR predictors of invasiveness and consistency of pituitary adenoma at the preoperative stage, with intraoperative confirmation. The Knosp Scale (1993) was used to assess laterosellar spread, and the Hardy and Vezina classification (1976) was used to assess suprasellar spread. Intraoperative assessment of pituitary adenoma consistency was carried out according to the classification of M. J. Rutkowski et al. (2020). Results. Among the surgical tactics selected for all patients, transsphenoidal endoscopic approach was used. Radical resection was achieved in 84 % of cases (N=84), subtotal — in 16 % (N=16). According to control MRI data 3‑6 months after surgical treatment, all subtotally operated patients showed continued tumor growth at different times, which subsequently required repeat transsphenoidal resection (in 14 %), and in 2 % — transcranial removal through a lateral supraorbital approach. No relationship was found with histological tumor subtypes, proliferative activity index ki-67 values, or immunohistochemical study results. Soft consistency of pituitary adenoma was predominantly encountered among the subjects (53 %), medium and dense consistency — less often (18 and 25 %, respectively). T1‑isointense MR signal is a predictor of intraoperatively softer density of pituitary adenoma (r=0.4; p=0.03). Hyperintense T2 signal in combination with increased PT and INR values indicates soft density of pituitary adenoma (p<0.001). Preoperative combination of laterosellar growth into the cavernous sinus (Knosp Scale 3‑4) and suprasellar spread (Hardy and Vezina 3‑4) indicates a high probability of subtotal tumor removal (p<0.05). Conclusions. Prediction of the fibrous residual supra- or laterosellar tumor fragment at the preoperative stage allows not only to determine the correct surgical approach, but also to determine the timing of follow-up examinations and the feasibility of radiosurgical treatment.

TL;DR: Neber et al. as discussed by the authors proposed a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently, which can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks.