Performance evaluation of the textural analysis algorithms in liver fibrosis detection using ultrasound software phantoms

Diagnosing liver fibrosis using non invasive procedures is challenging because the visual aspects in ultrasound imaging between healthy and fibrosis liver are very much alike. In this paper we use ultrasound simulation software to produce realistic liver tissue images. These images are produced by scanning several software liver phantoms. The phantoms proposed in this paper are generated based on relevant histological findings. The phantoms mimic the five METAVIR fibrosis stages from healthy liver to cirrhosis. A texture analysis based system, developed in our pervious studies is applied on these images. Accuracy in differentiating between healthy liver and various fibrosis stages is recorded. We show how various phantom properties affect the detection rates and present some explanations on why the liver fibrosis is not detectable to a level that is useful for physicians.