abstract
- In the continuous effort to mitigate the impacts of coronavirus disease (COVID-19) on global health, artificial intelligence (AI) has emerged as a promising ally, offering potential breakthroughs in the diagnosis and prognosis of lung diseases. Indeed, the use of supervised and unsupervised learning methods has the potential to aid in clinical decision-making and contribute to the comprehension of novel diseases. This study provides a comparative analysis of unsupervised and supervised methodolo-gies for accurately identifying ground-glass opacity (GGO) areas in CT scans. The GGO pulmonary lesion acts as a key diagnos-tic indicator of COVID-19 infection. Given the labor-intensive process of manually segmenting large chest CT datasets, there is an urgent requirement for dependable automated methods that facilitate efficient analysis of chest CT anatomy within extensive research databases. This need is particularly pronounced for less frequently annotated areas, such as pulmonary consolidations and radiological findings such as GGO lesions. To tackle this challenge, our study evaluates the performance of supervised and unsupervised learning methods using dice score, precision, and accuracy metrics. The evaluations are conducted on various datasets of annotated CT scans from COVID-19 patients. We consider that these findings are important in the context of COVID-19 diagnosis from CT scans and the relevance of both supervised and unsupervised learning techniques. Finally, we offer the open-source code of the experiments carried out in the research at https://github.com/gicc-lab/gicc_aimdai. © 2024 IEEE.