Modified U-Net for fully automatic liver segmentation from abdominal CT-image. (22nd August 2022)

• Record Type: Journal Article
• Authors: Mourya, Gajendra Kumar; Paul, Sudip; Handique, Akash; Baid, Ujjwal; Dutande, Prasad Vilas; Talbar, Sanjay Nilkant
• Is Part Of: International journal of biomedical engineering and technology. Volume 40:Number 1(2022)
• Page Start: 1
• Page End: 17
• Publication Date: 2022-08-22
• Subjects: computed tomography -- liver segmentation -- U-Net -- semantic segmentation -- deep learning; Biomedical engineering -- Periodicals; 610.28
• Languages: English
• ISSNs: 1752-6418

Abstract

Liver volume estimation using segmentation is the first step for liver diagnosis and its therapeutic planning. Liver segmentation from an abdominal CT image has always been a universal challenge for researchers because of low contrast among surrounding organs. An automatic liver segmentation technique is extremely desired in clinical practice. In this paper, we have modified conventional U-Net architecture for automatic liver segmentation. This method will precisely delineate the boundaries between the liver and other abdominal organs and outperforms over another state of the art methods. We extensively evaluated our method on 'CHAOS challenge-2019 dataset of 20 subjects' volumetric CT images. Quantitative evaluation of the proposed method is done in terms of various evaluation parameters with respect to their ground truth. Result achieved average dice similarity coefficient 0.97±0.03 and precision 0.93±0.12. In conclusion, the obtained results from this work demonstrated substantially significant performance with consistency and robustness.