A Novel and Efficient Digital Pathology Classifier for Predicting Cancer Biomarkers Using Sequencer Architecture.

Area of Science:

• Digital pathology
• Computational oncology
• Biomarker discovery

Background:

• Transformers and CNNs excel in digital pathology but are resource-intensive.
• There is a need for efficient and accurate digital pathology classifiers for cancer biomarker prediction.

Purpose of the Study:

• To develop and evaluate DPSeq, a novel and efficient digital pathology classifier.
• To predict key colorectal cancer biomarkers using histopathologic images.
• To compare DPSeq's performance against state-of-the-art CNN and transformer models.

Main Methods:

• Fine-tuned a sequencer architecture integrating bidirectional long short-term memory networks.
• Utilized hematoxylin and eosin-stained colorectal cancer images from The Cancer Genome Atlas and Molecular and Cellular Oncology datasets.
• Evaluated DPSeq's predictive performance for microsatellite instability, hypermutation, CpG island methylator phenotype, BRAF, TP53 mutations, and chromosomal instability.

Main Results:

• DPSeq demonstrated exceptional performance in predicting colorectal cancer biomarkers.
• Outperformed existing state-of-the-art classifiers in both internal and external cross-cohort validations.
• Achieved superior area under the receiver operating characteristic and precision-recall curves compared to four CNNs and two transformers for key biomarkers.
• Required less training and prediction time due to its simpler architecture.

Conclusions:

• DPSeq is a highly effective and efficient classifier for predicting colorectal cancer biomarkers.
• DPSeq surpasses transformer and CNN models in accuracy and resource efficiency.
• DPSeq represents a preferred alternative for biomarker prediction in digital pathology.