Ultrafast and Ultralight Network-Based Intelligent System for Real-time Diagnosis of Ear Diseases in Any Devices

• DOI: 10.48550/arxiv.2308.10610
• Publication Date: 2023-01-01
• Authors: Yubiao Yue; Xinyu Zeng; Xiaoqiang Shi; Meiping Zhang; Haihua Liang; Fan Zhang; Yanmei Chen; Zefeng Xie; Wenrui Wu; Zhenzhang Li
• Keywords: Feature (linguistics); Transfer of learning

Traditional ear disease diagnosis heavily depends on experienced specialists and specialized equipment, frequently resulting in misdiagnoses, treatment delays, financial burdens for some patients. Utilizing deep learning models efficient has proven effective affordable. However, existing research overlooked model inference speed parameter size required deployment. To tackle these challenges, we constructed a large-scale dataset comprising eight categories normal canal samples from two hospitals. Inspired by ShuffleNetV2, developed Best-EarNet, an ultrafast ultralight network enabling real-time diagnosis. Best-EarNet incorporates the novel Local-Global Spatial Feature Fusion Module which can capture global local spatial information simultaneously guide to focus crucial regions within feature maps at various levels, mitigating low accuracy issues. Moreover, our uses multiple auxiliary classification heads optimization. With 0.77M parameters, achieves average frames per second of 80 CPU. Employing transfer five-fold cross-validation with 22,581 images Hospital-1, impressive 95.23% accuracy. External testing 1,652 Hospital-2 validates its performance, yielding 92.14% Compared state-of-the-art networks, establishes new (SOTA) practical applications. Most importantly, intelligent system called Ear Keeper, be deployed common electronic devices. By manipulating compact otoscope, users perform comprehensive scanning using video. This study provides paradigm endoscopy other medical endoscopic image recognition