KRASFormer: a fully vision transformer-based framework for predictingKRASgene mutations in histopathological images of colorectal cancer
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.
View abstract on PubMed
Summary
This summary is machine-generated.A new AI framework, KRASFormer, predicts Kirsten Rat Sarcoma Virus (KRAS) gene mutations from standard H&E stained images. This offers a cost-effective and faster alternative to genetic sequencing for colorectal cancer treatment decisions.
Area Of Science
- Computational pathology
- Genomics
- Artificial intelligence in oncology
Background
- Kirsten Rat Sarcoma Virus (KRAS) gene mutations are critical biomarkers for colorectal cancer (CRC) treatment selection, particularly for anti-EGFR therapy.
- Current KRAS mutation detection methods like next-generation sequencing (NGS) are costly, time-consuming, and may not be universally applicable to all patient samples.
- There is a need for efficient and accessible methods to identify KRAS mutations to guide personalized treatment strategies in CRC.
Purpose Of The Study
- To develop and validate a novel computational framework, KRASFormer, for predicting KRAS gene mutations directly from Hematoxylin and Eosin (H&E) stained whole slide images (WSIs) of colorectal cancer.
- To assess the performance of a Vision Transformer-based XCiT model within KRASFormer for accurate KRAS mutation status prediction.
- To demonstrate the potential of H&E WSIs as a source for predicting KRAS mutations, offering a cost-effective and time-efficient alternative.
Main Methods
- Development of KRASFormer, a two-stage framework involving tumor region filtering and KRAS mutation prediction.
- Utilizing a quality screening mechanism in the first stage to isolate tumor cells from WSIs.
- Employing a Vision Transformer-based XCiT model in the second stage, leveraging cross-covariance attention for pattern recognition in tumor tissue.
Main Results
- The XCiT model achieved Area Under the ROC Curve (AUC) values of 0.691 on the TCGA-CRC-DX dataset and 0.653 on an in-house cohort for KRAS mutation prediction.
- The KRASFormer framework demonstrated superior performance compared to existing state-of-the-art methods in predicting KRAS mutations from WSIs.
- The initial stage of KRASFormer effectively filtered non-tumor regions, ensuring focus on relevant cellular material.
Conclusions
- H&E stained WSIs can be effectively utilized for predicting KRAS gene mutations in colorectal cancer, providing a cost-effective and time-efficient approach to guide treatment.
- Transformer-based models, like the XCiT employed, show significant potential for increasing performance metrics in computational pathology tasks.
- The successful development of KRASFormer highlights the value of interdisciplinary collaboration between pathologists and data scientists in creating clinically meaningful AI models.
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.
Related Concept Videos
The Ras-gene-encoded proteins are regulators of signaling pathways controlling cell proliferation, differentiation, or cell survival. The Ras-gene family in humans constitutes three primary members—the HRas, NRas, and KRas. These genes code for four functionally distinct yet closely related proteins—the HRas, NRas, KRas4A, and KRas4B. The involvement of mutant Ras genes in human cancer was first discovered in 1982 and is among the most common causes of human tumorigenesis.
Ras is a...
Tumor progression is a phenomenon where the pre-formed tumor acquires successive mutations to become clinically more aggressive and malignant. In the 1950s, Foulds first described the stepwise progression of cancer cells through successive stages.
Colon cancer is one of the best-documented examples of tumor progression. Early mutation in the APC gene in colon cells causes a small growth on the colon wall called a polyp. With time, this polyp grows into a benign, pre-cancerous tumor. Further...