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

Abstract

Background: Stimulated Raman histology (SRH) is a label-free optical imaging method for rapid intraoperative analysis of fresh tissue samples. Analysis of SRH images using Convolutional Neural Networks (CNN) has shown promising results for predicting the main histopathological classes of neurooncological tumors. Due to the relatively low number of rare tumor representations in CNN training datasets, a valid prediction of rarer entities remains limited. To develop new reliable analysis tools, larger datasets and greater tumor variety are crucial. One way to accomplish this is through research biobanks storing frozen tumor tissue samples. However, there is currently no data available regarding the pertinency of previously frozen tissue samples for SRH analysis. The aim of this study was to assess image quality and perform a comparative reliability analysis of artificial intelligence-based tumor classification using SRH in fresh and frozen tissue samples.

Methods: In a monocentric prospective study, tissue samples from 25 patients undergoing brain tumor resection were obtained. SRH was acquired in fresh and defrosted samples of the same specimen after varying storage durations at -80 °C. Image quality was rated by an experienced neuropathologist, and prediction of histopathological diagnosis was performed using two established CNNs.

Results: The image quality of SRH in fresh and defrosted tissue samples was high, with a mean image quality score of 1.96 (range 1-5) for both groups. CNN analysis showed high internal consistency for histo-(Cα 0.95) and molecular (Cα 0.83) pathological tumor classification. The results were confirmed using a dataset with samples from the local tumor biobank (Cα 0.91 and 0.53).

Conclusions: Our results showed that SRH appears comparably reliable in fresh and frozen tissue samples, enabling the integration of tumor biobank specimens to potentially improve the diagnostic range and reliability of CNN prediction tools.

Keywords: artificial intelligence; brain tumors; digital pathology; stimulated Raman histology.

Figure 1

Workflow of the test–retest analysis of the patient dataset. A small (3-4 mm) tissue sample (sample #1) was collected during surgery and immediately processed for SRH imaging. The fresh squash preparation was scanned in the SRH microscope (scan 1 sample #1) and frozen at −80 °C afterwards. After varying time intervals, the sample was defrosted and scanned again in the SRH microscope (re-scan sample #1). All SRH images from fresh and frozen samples were assessed for image quality and occurrence of freezing artifacts by an experienced neuropathologist and analyzed by the CNNs. CNN: Convolutional Neural Network.

Figure 2

SRH images of fresh and thawed tissue samples. Upper row: SRH images of a meningioma (CNS WHO grade 1) ((A): scan 1 sample #1, fresh; (B) re-scan 1, sample #1, defrosted), showing typical histologic features, such as meningothelial whorls (green arrows). Middle row: SRH images of a pulmonary adenocarcinoma metastasis ((C) scan 1, sample #1, fresh; (D) re-scan 1, sample #1, defrosted), showing sheets of epithelial tumor cells. Lower row: SRH images of a glioblastoma, IDH wild type (CNS WHO grade 4) ((E) scan 1, sample #1, fresh; (F) re-scan 1, sample #1, defrosted), showing infiltration of fibrillary tumor.

Figure 3

Confusion matrix of the CNN-based histological entity differentiation (left), and diffuse adult-type glioma subclassification (right) in fresh and frozen tumor tissue samples from the same patient. Ca = Cronbach’s alpha.

Figure 4

Confusion matrix of the CNN-based histological entity differentiation (left), and diffuse adult-type glioma subclassification (right) in tumor biobank samples comparing SRH images of fresh and frozen tumor samples. Ca = Cronbach’s alpha.