Medical Image Classification and Enhancement Using Machine Learning: A Focus on Fingerprint Colorized Data

Abstract

The consolidation of machine learning in medical image analysis has revolutionized diagnostic processes, specifically in the domain of patient identification and verification. Machine learning used in medical image analysis has transformed how patients are identified and verified for diagnosis. Fingerprint biometrics, which have historically been useful in forensic and civil identity applications, are now helping to secure patient authentication in healthcare. Even so, applying color to fingerprint data for medical use introduces new issues with accuracy and improving the quality of the images. The overall aim is to develop and test a sound computational system that not only enhances colorized image pattern recognition but also caters to the operational constraints of the medical environment. The dataset used in this study consisted of 35,000 structured fingerprint images that were synthetically colorized and labeled for identity classification tasks. They were drawn from several open-source and approved fingerprint repositories, including the NIST Special Database 302 and the Fingerprint Verification Competition (FVC) datasets, which were aligned with additional information using colorization algorithms developed for dermatoglyphic spectral analysis. This research project used three main model choices—ResNet, CNNs, and an MLP classifier—since they handled different strengths of the images we were working with. Each one of the three models—ResNet, CNN, and MLP—was trained and optimized using two main optimizers: Adam and SGD. An effective way of evaluating the models across several aspects was put together. How accurately the models were formed was the main measure of their performance. To assess the stability of the models, precision, recall, and F1-score were tallied for each class separately. The highest validation accuracy was attained by the ResNet18 model, suggesting that it did best on the test data compared to the others. Adding fingerprint biometric data to EHR systems considerably adds to the reliability and usefulness of the digital medical infrastructure. Because almost all medical providers now use certified EHR tools (as identified by the ONC), having secure and reliable login systems for each patient is more important than ever. Many interesting future approaches have the potential to address existing issues and improve what is known in the field. Applying GANs is one of the most interesting ways to produce realistic-looking fingerprint images.