| Creators: |
Hieber, Daniel and Prokop, Georg and Karthan, Maximilian and Märkl, Bruno and Schlegel, J. and Pryss, Rüdiger and Grambow, Gregor and Schobel, Johannes and Liesche-Starnecker, Friederike |
| Title: |
Machine learning-based assessment of intratumor heterogeneity in glioblastoma |
| Item Type: |
Conference or Workshop Item |
| Event Title: |
(Abstracts of the) 20th International Congress of Neuropathology (ICN) |
| Event Location: |
Berlin, Germany |
| Event Dates: |
September, 13-16, 2023 |
| Projects: |
DigiHealth, NAP |
| Page Range: |
Abstr. eP-NO-A78 |
| Additional Information: |
Special Issue: Abstracts of the 20th International Congress of Neuropathology, Berlin, Germany, September 13–16, 2023. Brain Pathology, 33: e13194. https://doi.org/10.1111/bpa.13194 |
| Date: |
2023 |
| Divisions: |
Gesundheitsmanagement |
| Abstract (ENG): |
Introduction: Currently, there exist no standardized approaches to objectively and reproducibly determine the intratumor hetereogeneity of glioblastoma (GBM). Instead, neuropathologists manually analyze GBM in a time-consuming and error-prone process with no guarantee of exact reproducibility.
Objectives: to objectively determine the heterogeneity of GBM, a machine learning (ML) model is trained based on whole slide images (SI) of hematoxylin and eosin (HE) stained slides. The model shall be able to reproducibly determine the heterogeneity of GBM without any additional input.
Material and Methods: Based on 103 GBM HE-stained WSI from 56 patients, a classification ML model was developed and trained to assess the degree of heterogeneity. Out of these 103 images, 38 were selected as basis for the trained model. Using a newly implemented image analysis algorithm, each WSI was converted into a smaller image showing only four representative regions of the tumor issue, distributed throughout the original image. Using nested cross-validation, an optimal ML model with hyperparameters was selected to perform the classification task.
Results: Based on the available 38 input images, a heterogeneity classification accuracy of approx. 67% was achieved. While this may not seem like a huge success, values above 60% are generally considered a working classification. Given the small sample size (i.e., only 38 images) this can be seen as a strong validation of the feasibility and a very promising approach. Including pre-processing of the WSI, the complete analysis can be processed in less than 3 min. per image.
Conclusions: Based on 38 HE stained WSI, an ML model was trained for the classification of intratumor heterogeneity in GBM. Considering the small amount of available data, the achieved accuracy of 67% can be considered a great success and strongly validates the feasibility of further research. |
| Forthcoming: |
No |
| Language: |
English |
| Link eMedia: |
Download |
| Citation: |
Hieber, Daniel and Prokop, Georg and Karthan, Maximilian and Märkl, Bruno and Schlegel, J. and Pryss, Rüdiger and Grambow, Gregor and Schobel, Johannes and Liesche-Starnecker, Friederike
(2023)
Machine learning-based assessment of intratumor heterogeneity in glioblastoma.
In: (Abstracts of the) 20th International Congress of Neuropathology (ICN), September, 13-16, 2023, Berlin, Germany, Abstr. eP-NO-A78.
|
