Home / Current Issue / Paper 1716822
AI-Assisted Early Detection of Oral Cancer Using Attention-Guided Deep Learning and Automated Diagnostic Reporting
Subject area: Science,Engineering and Technology · Area of research: Artificial Intelligence in Healthcare
DOI: https://doi.org/10.64388/IREV9I10-1716822
Abstract
Late-stage oral cancer diagnosis is a problem we have had the tools to address for years — the bottleneck has always been getting those tools into the right hands fast enough. Our work grew out of that frustration. We put together AGMSFFNet, a histopathology classifier that pairs a modified EfficientNet-B5 backbone with a dual-branch attention design we call HSCA, which handles both spatial and channel recalibration at once. Alongside that, multi-resolution LBP texture descriptors feed into the feature pipeline to capture detail that convolution alone tends to skip over. The part we are most invested in, though, is the LangGraph reporting layer — rather than stopping at a class label, the system drafts a structured clinical summary automatically, flags anything it is uncertain about, and hands a readable report to the pathologist. Testing on ORCHID gave us 98.7% accuracy, 98.9% precision, 98.5% recall, and a 98.7% F1. External results on NDBUFES held up at 97.5%. ANOVA and Tukey tests confirmed the differences were statistically meaningful.
Keywords
Oral Cancer Classification, Deep Learning, Efficientnet, Attention Mechanisms, Feature Fusion, Histopathological Image Analysis, Langgraph, Agentic AI, Interpretable AI.
How to cite this paper
@article{1716822,
author = {Sneha Sawant, Lokesh Thakare, Mushfiq Shaikh, Suhas Waghmare},
title = {AI-Assisted Early Detection of Oral Cancer Using Attention-Guided Deep Learning and Automated Diagnostic Reporting},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {9},
number = {10},
pages = {2577-2580},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1716822.pdf},
abstract = {Late-stage oral cancer diagnosis is a problem we have had the tools to address for years — the bottleneck has always been getting those tools into the right hands fast enough. Our work grew out of that frustration. We put together AGMSFFNet, a histopathology classifier that pairs a modified EfficientNet-B5 backbone with a dual-branch attention design we call HSCA, which handles both spatial and channel recalibration at once. Alongside that, multi-resolution LBP texture descriptors feed into the feature pipeline to capture detail that convolution alone tends to skip over. The part we are most invested in, though, is the LangGraph reporting layer — rather than stopping at a class label, the system drafts a structured clinical summary automatically, flags anything it is uncertain about, and hands a readable report to the pathologist. Testing on ORCHID gave us 98.7% accuracy, 98.9% precision, 98.5% recall, and a 98.7% F1. External results on NDBUFES held up at 97.5%. ANOVA and Tukey tests confirmed the differences were statistically meaningful.},
keywords = {Oral Cancer Classification, Deep Learning, Efficientnet, Attention Mechanisms, Feature Fusion, Histopathological Image Analysis, Langgraph, Agentic AI, Interpretable AI.},
month = {April},
doi = {https://doi.org/10.64388/IREV9I10-1716822}
}