Communications Medicine (Mar 2024)

Development and prospective validation of an artificial intelligence-based smartphone app for rapid intraoperative pituitary adenoma identification

  • Rabih Bou-Nassif,
  • Anne S. Reiner,
  • Matthew Pease,
  • Tejus Bale,
  • Marc A. Cohen,
  • Marc Rosenblum,
  • Viviane Tabar

DOI
https://doi.org/10.1038/s43856-024-00469-z
Journal volume & issue
Vol. 4, no. 1
pp. 1 – 7

Abstract

Read online

Abstract Background Intraoperative pathology consultation plays a crucial role in tumor surgery. The ability to accurately and rapidly distinguish tumor from normal tissue can greatly impact intraoperative surgical oncology management. However, this is dependent on the availability of a specialized pathologist for a reliable diagnosis. We developed and prospectively validated an artificial intelligence-based smartphone app capable of differentiating between pituitary adenoma and normal pituitary gland using stimulated Raman histology, almost instantly. Methods The study consisted of three parts. After data collection (part 1) and development of a deep learning-based smartphone app (part 2), we conducted a prospective study that included 40 consecutive patients with 194 samples to evaluate the app in real-time in a surgical setting (part 3). The smartphone app’s sensitivity, specificity, positive predictive value, and negative predictive value were evaluated by comparing the diagnosis rendered by the app to the ground-truth diagnosis set by a neuropathologist. Results The app exhibits a sensitivity of 96.1% (95% CI: 89.9–99.0%), specificity of 92.7% (95% CI: 74–99.3%), positive predictive value of 98% (95% CI: 92.2–99.8%), and negative predictive value of 86.4% (95% CI: 66.2–96.8%). An external validation of the smartphone app on 40 different adenoma tumors and a total of 191 scanned SRH specimens from a public database shows a sensitivity of 93.7% (95% CI: 89.3–96.7%). Conclusions The app can be readily expanded and repurposed to work on different types of tumors and optical images. Rapid recognition of normal versus tumor tissue during surgery may contribute to improved intraoperative surgical management and oncologic outcomes. In addition to the accelerated pathological assessments during surgery, this platform can be of great benefit in community hospitals and developing countries, where immediate access to a specialized pathologist during surgery is limited.