Zhang Xu, Xu Wei, Xu Zheng, Tong Henry H Y, Jiao Xueping, Li Kefeng, Wang Zhiwen
School of Nursing, Peking University, Beijing, People's Republic of China.
Faculty of Applied Sciences, Macao Polytechnic University, Macao, People's Republic of China.
J Multidiscip Healthc. 2025 Apr 18;18:2215-2223. doi: 10.2147/JMDH.S515644. eCollection 2025.
Irritant dermatitis is a common complication among stoma patients, significantly impacting their quality of life. Early diagnosis is essential, but limited access to healthcare and poor self-management skills often delay treatment. This study aimed to assess the effectiveness of two advanced convolutional neural networks (CNNs), ConvNeXt and MobileViT, for the intelligent diagnosis of irritant dermatitis using smartphone-acquired stoma images.
A retrospective observational study was conducted, collecting 825 stoma complication images from five tertiary hospitals in China. Data preprocessing techniques such as resampling and enhancement were used to prepare the dataset. The ConvNeXt and MobileViT models were trained and evaluated based on accuracy, precision, recall, and F1 scores. Optimizers and learning rates were also adjusted to assess model performance.
ConvNeXt demonstrated superior performance, achieving an accuracy of 71.4%, precision of 73.6%, recall of 67.1%, and an F1 score of 70.2% with the Adam optimizer and a 0.001 learning rate. MobileViT, despite being more lightweight, did not surpass ConvNeXt, with a maximum accuracy of 64.4%. ConvNeXt excelled in diagnosing irritant dermatitis and normal stoma conditions but showed limitations in recognizing other complications.
The ConvNeXt model outperformed MobileViT, indicating that advanced CNNs can effectively assist in the early diagnosis of irritant dermatitis among stoma patients. This could help alleviate the burden on healthcare resources and improve patient outcomes through accessible mobile-based diagnostic tools.
刺激性皮炎是造口患者常见的并发症，严重影响他们的生活质量。早期诊断至关重要，但医疗服务获取受限和自我管理技能较差常常延误治疗。本研究旨在评估两种先进的卷积神经网络（ConvNeXt和MobileViT）利用智能手机获取的造口图像对刺激性皮炎进行智能诊断的有效性。
开展一项回顾性观察研究，从中国五家三级医院收集825张造口并发症图像。使用重采样和增强等数据预处理技术来准备数据集。基于准确率、精确率、召回率和F1分数对ConvNeXt和MobileViT模型进行训练和评估。还调整了优化器和学习率以评估模型性能。
ConvNeXt表现出卓越的性能，在使用Adam优化器和0.001学习率时，准确率达到71.4%，精确率为73.6%，召回率为67.1%，F1分数为70.2%。MobileViT尽管更轻量级，但未超过ConvNeXt，最高准确率为64.4%。ConvNeXt在诊断刺激性皮炎和正常造口状况方面表现出色，但在识别其他并发症方面存在局限性。
ConvNeXt模型优于MobileViT，表明先进的卷积神经网络可以有效辅助造口患者刺激性皮炎的早期诊断。这有助于减轻医疗资源负担，并通过便捷的基于移动设备的诊断工具改善患者预后。
