Survey on Approaches to Deal with Limited Dataset for Liver Lesion Classification Using Deep Learning

Abstract

The one of the prominent cause of death in all over the world is Liver cancer. Various medical imaging techniques such as X-rays, Computed tomography (CT), ultrasound (US), magnetic resonance imaging (MRI) are now available for detection and categorization of liver tumor. The manually detection of liver tumor is a challenging task and time intensive. There are plenty of works has been done using Computer aided diagnosis (CAD) systems based on these medical images with deep learning which enhanced the liver tumor detection performance ratio. Due to non-invasive and cost effectiveness, CT image is widely used in liver tumor detection. The one of the major advantage of deep learning is that they extracts low and high level features from underlying data automatically rather than feeding handcrafted feature extraction techniques. Although there are several issue with deep learning. One of the barrier with deep learning is the requirement of large annotated data set but in medical field, mostly the available annotated image data set is limited. The objective of this paper to provide an overview of various methods to cope with this barrier especially in liver lesion detection and classification.