| Creators: |
Dinser, Moritz and Hieber, Daniel and Pryss, Rüdiger and Monoranu, Camelia-Maria and Liesche-Starnecker, Friederike and Schobel, Johannes and Nickl, Vera |
| Title: |
Automated Locoregional Antigen Determination as a Prerequisite for CAR T-Cell Therapy in Glioblastoma (Automatisierte lokoregionale Antigenbestimmung als Voraussetzung für eine CAR T-Zell Therapie bei Glioblastom) |
| Item Type: |
Conference or Workshop Item |
| Event Title: |
Sektionstage der deutschen Gesellschaft für Neurochirurgie (ST Herbst 2024) |
| Event Location: |
Würzburg, Germany |
| Event Dates: |
19.-21. September 2024 |
| Projects: |
DigiHealth, NAP |
| Date: |
2024 |
| Divisions: |
Gesundheitsmanagement |
| Abstract: |
Objective:
Glioblastoma (GBM) is the most malignant brain tumor in adults, with a median survival rate of 15-21 months under standard therapy. Clinical studies show that T-cells can be modified to express chimeric antigen receptors (CARs) and can be retargeted to recognize tumor-associated antigens (TAAs). However, selecting suitable TAAs for GBM proves
challenging as it shows highly heterogeneous intra- and intertumoral TAA expression patterns, increasing the risk for antigen escape and relapse. To avoid the latter, comprehensive coverage across the tumor must be ensured while minimizing target antigen overlap to reduce the risk of off-tumor toxicity. This work presents an automated artificial intelligence pipeline that determines TAAs with maximum coverage and minimal overlap based on immunohistochemical (IHC) stained pathological slides in GBM as a prerequisite to choosing the optimal therapeutic CAR T-cell combination.
Methods:
The proposed pipeline utilizes one hematoxylin and eosin (HE) and six IHC-stained slides for different TAAs expressed in GBM. Following the initial alignment of all slides to ensure congruent tissue distribution, a machine learning (ML) model identifies the tumor regions in the HE slide and creates a tumor mask. Based on the mask, tumor regions from
the IHC slides are extracted and segmented into smaller tiles. Next, a color deconvolution algorithm is applied, and each tile's antigen coverage is quantified. Finally, antigen expressions from all IHC slides are combined using pixel-wise comparison to determine antigen pairs or triplets with minimal overlap and maximum coverage.
Results:
The proposed pipeline is currently evaluated in a study comparing the results to previous conventional analyses,
determining optimal TAAs for a dual or triple CAR T-cell therapy strategy. The study evaluates six GBM-specific TAAs.
The first results align with the results of the conventional analyses.
Conclusion:
This work presents a novel automated TAA target selection pipeline for multitargeted CAR T-cell therapy in GBM patients. The initial results are promising and consistent with accompanying analyses. The pipeline uses a minimum of ML, limited to tumor detection, ensuring easy-to-understand results. It may serve clinicians as a decision support
system for CAR T-cell therapy target selection. For future research, validation of the pipeline"s suitability for TAA selection is planned using organoid models. |
| Forthcoming: |
No |
| Language: |
English |
| Citation: |
Dinser, Moritz and Hieber, Daniel and Pryss, Rüdiger and Monoranu, Camelia-Maria and Liesche-Starnecker, Friederike and Schobel, Johannes and Nickl, Vera
(2024)
Automated Locoregional Antigen Determination as a Prerequisite for CAR T-Cell Therapy in Glioblastoma (Automatisierte lokoregionale Antigenbestimmung als Voraussetzung für eine CAR T-Zell Therapie bei Glioblastom).
In: Sektionstage der deutschen Gesellschaft für Neurochirurgie (ST Herbst 2024), 19.-21. September 2024, Würzburg, Germany.
|
