Pixel-level DNA encoding with codon pattern feature extraction for oral cancer classification using transformer ensembles

Oral cancer is among the most aggressive cancer types worldwide, and early diagnosis and accurate classification are important for improving diagnosis and patient prognosis. Many oral cancer cases are diagnosed at advanced stages, whereas detection at precancerous stages can contribute to reducing mortality. Therefore, there is a need for effective image-based approaches capable of providing accurate and reliable oral cancer classification. This study proposes a pixel-level DNA encoding framework with codon-based feature extraction for oral cancer image classification. Each image is transformed into DNA planes using an encoding scheme that considers the first six bits of each pixel and converts them into nucleotide sequences while maintaining run-length and GC-content constraints. Three consecutive nucleotides are grouped to form codons, and probability histograms of the codon patterns are computed to construct feature vectors. These feature vectors are classified using TabPFN, TabTransformer, SAINT, and FT-Transformer, together with an ensemble model. The framework was evaluated on ten publicly accessible datasets comprising four clinical and six histopathological image datasets. On the clinical datasets, the proposed framework achieved 98.90% precision, 98.89% recall, 98.90% F1-score, and 98.90% accuracy. On the histopathological datasets, it achieved 97.49% precision, 98.09% recall, 97.78% F1-score, and 97.99% accuracy. The results demonstrate that the proposed pixel-level DNA encoding and codon-based feature extraction framework provides a robust and accurate approach for oral cancer image classification. The compact codon-derived representation combined with transformer-based ensemble classification shows its potential for analyzing both clinical and histopathological oral cancer images.

Authors

Institutions

Publication Details

Journal
BMC Bioinformatics
Published
2026-11-18
DOI
https://doi.org/10.1186/s12859-026-06685-9
Primary Topic
AI in cancer detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Pixel-level DNA encoding with codon pattern feature extraction for oral cancer classification using transformer ensembles

Govind Narayan Patel, Jitesh Pradhan, Tapan Kumar Dey
1 links
BMC Bioinformatics
AI in cancer detection
article

Pixel-level DNA encoding with codon pattern feature extraction for oral cancer classification using transformer ensembles

Govind Narayan Patel, Jitesh Pradhan, Tapan Kumar Dey
article en

Abstract

Oral cancer is among the most aggressive cancer types worldwide, and early diagnosis and accurate classification are important for improving diagnosis and patient prognosis. Many oral cancer cases are diagnosed at advanced stages, whereas detection at precancerous stages can contribute to reducing mortality. Therefore, there is a need for effective image-based approaches capable of providing accurate and reliable oral cancer classification. This study proposes a pixel-level DNA encoding framework with codon-based feature extraction for oral cancer image classification. Each image is transformed into DNA planes using an encoding scheme that considers the first six bits of each pixel and converts them into nucleotide sequences while maintaining run-length and GC-content constraints. Three consecutive nucleotides are grouped to form codons, and probability histograms of the codon patterns are computed to construct feature vectors. These feature vectors are classified using TabPFN, TabTransformer, SAINT, and FT-Transformer, together with an ensemble model. The framework was evaluated on ten publicly accessible datasets comprising four clinical and six histopathological image datasets. On the clinical datasets, the proposed framework achieved 98.90% precision, 98.89% recall, 98.90% F1-score, and 98.90% accuracy. On the histopathological datasets, it achieved 97.49% precision, 98.09% recall, 97.78% F1-score, and 97.99% accuracy. The results demonstrate that the proposed pixel-level DNA encoding and codon-based feature extraction framework provides a robust and accurate approach for oral cancer image classification. The compact codon-derived representation combined with transformer-based ensemble classification shows its potential for analyzing both clinical and histopathological oral cancer images.

BMC Bioinformatics
National Institute of Technology Jamshedpur (IN), Manipal University Jaipur (IN)
AI in cancer detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon