Automated (artificial intelligence) vs. manual evaluation of the somatostatin-receptor 2A and proliferation marker Ki-67 in neuroendocrine neoplasms: A pilot study

#4281

Introduction: Digital image analysis methods are currently being equipped with artificial intelligence (AI). Various AI applications are already in use to determine Ki-67. To date, there is no specialised AI application available to determine the somatostatin-receptor 2A (SSTR2A).

Aim(s): Evaluation of the automated vs. manual analysis and validation of AI algorithms usability by experts.

Materials and methods: Archived neuroendocrine neoplasms (NEN; n = 278) were sectioned and immunohistochemical staining was carried out to determine Ki-67 and SSTR2A. The slides (n = 300) were then digitised and evaluated for Ki-67 and SSTR2A using two AI algorithms (Mindpeak GmbH, Hamburg, Germany). In parallel, a conventional manual evaluation was carried out by 3 independent endocrine pathologists (reference). In part 2, 9 histopathological experts were surveyed in an online study after testing the AI on usability using a standardised questionnaire (system-usability-scale SUS).

Conference:

Presenting Author:

Authors: Kaemmerer D, Lupp A, Klöppel G, Ayako I, Kasajima A,

Keywords: artificial intelligence, Ki-67, SSTR2A, usability, neuroendocrine neoplasm,