JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Image Quality Assessment and Reliability Analysis of Artificial Intelligence-Based Tumor Classification of Stimulated Raman Histology of Tumor Biobank Samples

  • 0Department of General Neurosurgery, Center for Neurosurgery, Faculty of Medicine and University Hospital Cologne, University of Cologne, 50937 Cologne, Germany.
Diagnostics (basel, Switzerland) +

|

December 17, 2024

|

December 17, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

Stimulated Raman histology (SRH) reliably analyzes fresh and frozen brain tumor tissues. This enables using biobank samples to enhance artificial intelligence diagnostic tools for rare tumors.

Area Of Science

  • Biomedical Optics
  • Artificial Intelligence in Pathology
  • Neuro-oncology

Background

  • Stimulated Raman histology (SRH) is a label-free imaging technique for intraoperative tissue analysis.
  • Convolutional Neural Networks (CNNs) show promise in classifying neuro-oncological tumors from SRH images.
  • Limited data exists on SRH analysis of frozen tissues, hindering the use of biobank samples for expanding CNN training datasets.

Purpose Of The Study

  • To assess the image quality of SRH in fresh versus frozen brain tumor tissues.
  • To compare the reliability of AI-based tumor classification using SRH on fresh and frozen samples.
  • To evaluate the potential of integrating biobank specimens into SRH-based AI diagnostic tools.

Main Methods

  • Prospective study involving 25 brain tumor resection specimens.
  • SRH imaging performed on fresh and cryopreserved (-80°C) samples.
  • Image quality assessed by a neuropathologist; tumor classification performed using two established CNNs.

Main Results

  • High image quality observed for both fresh and frozen SRH samples (mean score 1.96/5).
  • CNNs demonstrated high consistency in histopathological (Cα 0.95) and molecular (Cα 0.83) tumor classification.
  • Results were validated using local tumor biobank samples, showing good performance (Cα 0.91 and 0.53).

Conclusions

  • SRH provides comparable reliability for both fresh and frozen tissue samples.
  • Frozen tissue samples from biobanks can be integrated with SRH analysis.
  • This integration has the potential to broaden the diagnostic scope and improve the accuracy of AI-based CNN prediction tools for neuro-oncological tumors.

Keywords: