LAVIA-MXID: A New Annotated X-Ray Medical Image Dataset

X-ray medical imaging datasets are crucial for accurate diagnosis and driving innovation in medical imaging technologies. The LAVIA-MXID extended dataset, curated by the Laboratory of Vision and Artificial Intelligence (LAVIA), addresses limitations of its predecessor, the MXID dataset. It offers increased diversity, nearly 10,000 annotated images, and features tailored for advanced medical imaging applications. This study presents a handcrafted machine learning and deep learning approach for medical image compression, focusing on recognizing peculiarities and problems in medical imaging pattern recognition, such as data set quality, labeling errors, and data imbalance. The LAVIA-MXID dataset, an enhanced version of the MXID dataset, features 9989 images managed by radiologists and medical experts. It is meticulously labeled, categorized into training, validation, and test subsets, and isolated and eliminated for high-quality annotation standards. It supports various image analysis tasks but does not include annotations for specific pathological conditions. The study analyzed 9,989 medical images from the LAVIA-MXID dataset using machine learning and deep learning methods. The results showed marginal improvements in the extended dataset (version 2), but enhanced image quality with reduced reconstruction artifacts. The four models - Autoencoder (AE), Deep Convolutional Autoencoder (DCAE), Convolutional Neural Network (CNN), and Variational Autoencoder (VAE), showed improvement in performance metrics like MSE, PSNR, and MS-SSIM. The proposed dataset and methods improve X-ray medical imaging diagnostic accuracy by training machine learning algorithms, addressing issues like dataset size, quality, labeling errors, and data imbalance. The LAVIA-MXID dataset, developed by LAVIA, offers advanced medical image analysis using machine learning and deep learning paradigms, enhancing compression quality.