Positional encoding cross-attention ResNet50 framework for lung cancer classification across CT and histopathological imaging modalities

Lung cancer remains one of the leading causes of cancer-related mortality worldwide, primarily due to delayed diagnosis and limitations associated with manual interpretation of medical images. Deep learning approaches, particularly convolutional neural networks such as ResNet50, have demonstrated significant potential for automated medical image analysis. However, accurately capturing both low-level structural information and high-level semantic information within medical images remains a challenging task. This study proposes a Positional Encoding Cross-Attention ResNet50 framework that enhances feature representation learning through interactions between shallow structural features and deep semantic features extracted from different stages of the same backbone network. A two-dimensional positional encoding mechanism is incorporated to preserve spatial information, while a cross-attention module enables the model to selectively emphasize diagnostically relevant image regions. The proposed framework was independently evaluated on publicly available chest CT scan and histopathological image datasets for lung cancer classification. Experimental results demonstrate that the attention-enhanced architecture improves classification performance by effectively integrating structural and semantic information within each imaging modality. In addition, a separate rule-based clinical decision-support module provides indicative assessments of cancer stage, severity, and recovery range based on externally supplied clinical parameters. These outputs should be interpreted as decision-support estimates rather than validated prognostic predictions. The findings demonstrate the effectiveness of intra-modality attention-guided feature learning for lung cancer image classification across multiple imaging modalities.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-12
DOI
https://doi.org/10.1038/s41598-026-77608-0
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

Positional encoding cross-attention ResNet50 framework for lung cancer classification across CT and histopathological imaging modalities

A. Nivashini, M. Krishnamurthy
Scientific Reports
AI in cancer detection
article

Positional encoding cross-attention ResNet50 framework for lung cancer classification across CT and histopathological imaging modalities

A. Nivashini, M. Krishnamurthy
article en

Abstract

Lung cancer remains one of the leading causes of cancer-related mortality worldwide, primarily due to delayed diagnosis and limitations associated with manual interpretation of medical images. Deep learning approaches, particularly convolutional neural networks such as ResNet50, have demonstrated significant potential for automated medical image analysis. However, accurately capturing both low-level structural information and high-level semantic information within medical images remains a challenging task. This study proposes a Positional Encoding Cross-Attention ResNet50 framework that enhances feature representation learning through interactions between shallow structural features and deep semantic features extracted from different stages of the same backbone network. A two-dimensional positional encoding mechanism is incorporated to preserve spatial information, while a cross-attention module enables the model to selectively emphasize diagnostically relevant image regions. The proposed framework was independently evaluated on publicly available chest CT scan and histopathological image datasets for lung cancer classification. Experimental results demonstrate that the attention-enhanced architecture improves classification performance by effectively integrating structural and semantic information within each imaging modality. In addition, a separate rule-based clinical decision-support module provides indicative assessments of cancer stage, severity, and recovery range based on externally supplied clinical parameters. These outputs should be interpreted as decision-support estimates rather than validated prognostic predictions. The findings demonstrate the effectiveness of intra-modality attention-guided feature learning for lung cancer image classification across multiple imaging modalities.

Scientific ReportsVol. 16(1)
Anna University, Chennai (IN), KCG College of Technology (IN)
Good health and well-being
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