Analysis and Classification of Neonatal Lung Ultrasound Images

Annually 8,500 neonatal deaths are reported in the US due to respiratory failure, which is 1/3rd of all neonatal deaths in the US. Lung ultrasound (LUS) is easy to use, cost-effective, ionizing radiation-free, real-time imaging technique that can be used to diagnose and monitor neonatal lung pathologies. An automated screening system that captures characteristics of the LUS patterns can be of significant assistance to clinicians, especially in rural areas. In this thesis, one such system is presented using two approaches: (A1) using simple recurrence patterns present in the scanlines of the images and (A2) using wavelet image decomposition method. Using these two approaches, with 18 features, a leave-one-out image classification accuracy as high as 97.64% (A1) and 99.03% (A2) were achieved with a balanced database of 24 neonates and 92.77% (A1) and 94.06% (A2) were achieved with the whole database of 42 neonates.