Shavkatovich Buriboev Abror, Abduvaitov Akmal, Jeon Heung Seok
Department of AI-Software, Gachon University, Seongnam-si 13120, Republic of Korea.
Department of IT, Samarkand Branch of Tashkent University of Information Technologies, Samarkand 140100, Uzbekistan.
Sensors (Basel). 2025 Jun 26;25(13):3976. doi: 10.3390/s25133976.
Pneumonia remains a critical health concern, necessitating accurate and automated diagnostic tools. This study proposes a novel approach for the binary classification of pneumonia in chest X-ray images using an adaptive contrast enhancement model and a convolutional neural network (CNN). The enhancement model, an improvement over standard contrast-limited techniques, employs adaptive tile sizing, variance-guided clipping and entropy-weighted redistribution to optimize image quality for pneumonia detection. Applied to the Chest X-Ray Images (Pneumonia) dataset (5856 images), the enhanced images enable the CNN to achieve an accuracy of 98.7%, precision of 99.3%, recall of 98.6% and F1-score of 97.9%, outperforming baseline methods. The model's robustness is validated through five-fold cross-validation, and its feature extraction is visualized to ensure clinical relevance. Limitations, such as reliance on a single dataset, are discussed, with future evaluations planned for larger datasets like CheXpert and NIH Chest X-ray to enhance generalizability. This approach demonstrates the potential of tailored preprocessing and efficient CNNs for reliable pneumonia classification, contributing to improved diagnostic support in medical imaging.
肺炎仍然是一个严重的健康问题，因此需要准确且自动化的诊断工具。本研究提出了一种新颖的方法，利用自适应对比度增强模型和卷积神经网络（CNN）对胸部X光图像中的肺炎进行二元分类。该增强模型是对标准对比度受限技术的改进，采用自适应切片大小调整、方差引导裁剪和熵加权重新分配来优化用于肺炎检测的图像质量。将其应用于胸部X光图像（肺炎）数据集（5856张图像），增强后的图像使CNN能够达到98.7%的准确率、99.3%的精确率、98.6%的召回率和97.9%的F1分数，优于基线方法。通过五折交叉验证验证了该模型的稳健性，并对其特征提取进行了可视化以确保临床相关性。讨论了诸如依赖单一数据集等局限性，并计划对CheXpert和NIH胸部X光等更大的数据集进行未来评估，以提高通用性。这种方法展示了定制预处理和高效CNN在可靠肺炎分类方面的潜力，有助于改善医学成像中的诊断支持。
