EXPLORING IMAGING MODALITIES USING MACHINE LEARNING FOR LUNG DISEASE DIAGNOSIS

Lung disorders, both infectious and noninfectious, are a primary cause of death worldwide. Pneumonia, lung cancer, and COVID-19 have emerged as particularly important. This paper presents a detailed investigation of machine learning algorithms for identifying these common lung diseases using multiple imaging modalities. Our review includes a variety of diagnostic imaging modalities, including X-rays, CT scans, MRI, PET scans, and specialized techniques, and assesses their benefits, limits, and uses in lung disease diagnosis. We thoroughly examine how these imaging datasets are used as basic inputs for machine learning-based diagnostic systems. We analyze current machine learning paradigms and clinical applications by conducting an exhaustive review of peer-reviewed literature from key academic databases (ScienceDirect, arXiv, IEEE Xplore, MDPI, and others). Transfer learning and ensemble learning methodologies improve CNN performance even further. While accuracy is still the most often used assessment criterion, our study finds considerable standardization, with issues in dataset many collections suffering from class imbalance and low diversity. The analysis emphasizes potentially developing technologies, such as explainable AI and federated learning, that have the potential for therapeutic use. Furthermore, the combination of multimodal imaging and patient information outperforms single modality techniques. We conclude by suggesting future research approaches, highlighting the importance of bigger, more diverse datasets and standardized assessment criteria that better represent clinical value than basic accuracy scores.