Deep-learning research for early laryngeal cancer detection from narrow-band imaging — going beyond binary diagnosis to eight clinically distinct classes. Published in Springer, 2026, and covered across national media.
Where existing systems classify tissue as merely benign or malignant, our model distinguishes eight distinct laryngeal conditions — enabling differential diagnosis at a granularity not possible before.
The most aggressive malignant class.
Non-invasive carcinoma — a critical early-intervention target.
Mild dysplasia — the earliest precancerous change.
Moderate dysplasia — intermediate risk, needs monitoring.
Severe dysplasia — a high-risk precancerous condition.
Benign laryngeal papilloma — viral aetiology, recurrent.
Benign keratotic lesion — distinguished from early dysplasia.
Normal laryngeal mucosa — the baseline reference class.
Accurately distinguishing between laryngeal diseases is a difficult undertaking that frequently leads to diagnostic mistakes due to expert subjectivity. Most existing automated systems are limited to a binary classification of benign versus malignant tissue.
This study introduces a deep-learning framework performing fine-grained classification of Contact Endoscopy–Narrow Band Imaging (CE-NBI) into eight distinct classes, including squamous cell carcinoma, various grades of dysplasia, and papillomatosis. A U-shaped network augmented with a specialised attention block selectively focuses on the most discriminative features within endoscopic images.
The approach outperforms state-of-the-art techniques, and its efficient design makes it compatible with low-memory hardware — enabling deployment in district hospitals and resource-constrained clinical settings across India.
Book: Data Science and Security
Series: LNNS, Vol. 1945
Publisher: Springer, Cham
Date: 26 May 2026
Dr. Rakesh Srivastava, Chief Consultant, ENT Dept., Health City Vistaar Hospital, Gomtinagar.
Covered across leading English and Hindi publications.
The research featured amid a university AI-in-education push, with new AI programmes launching from the 2026–27 session.
An early laryngeal-cancer detection AI developed by the research team's computer-science group.
Highlighted the clinical promise of NBI-based AI for early, biopsy-free diagnosis and treatment guidance.
Narrow-band imaging methodology covered in detail — the model identifies eight distinct laryngeal conditions.
Coverage of clinical applicability — enabling timely treatment through precise early-stage identification.
Coverage of academic mentorship programmes and PrajniX Labs' research partnerships.
Current capabilities available for institutional and clinical collaboration.
Attention-augmented U-Net validated on CE-NBI data, extensible to colonoscopy and bronchoscopy imaging.
Fine-grained frameworks enabling differential diagnosis — beyond the binary benign/malignant limit of prior systems.
NLP over radiology reports and clinical records — extracting structured insight for decision support.
Architectures for low-memory clinical hardware — putting advanced diagnostics within reach of district hospitals.
We're seeking clinical collaborators, hospital partnerships, and research co-authors for multi-cancer screening and real-time surgical guidance.