Abstract
Fatty liver screening is possible through the analysis of ultrasound -noninvasive and publicly available- images, but due to the complexity, skill of specialist directly effects the accuracy of results. An integrated quantitative index (FLDI) based on radon and cosine transform features of the ultrasound liver image in the literature claimed perfectly identifying fatty liver disease. Current study challenges the index validation, and examines the index using two other databases. Unfortunately no significant difference found between normal and fatty liver indices in new datasets; and normal index variation found significantly different among three datasets (1.111E6 ± 1.079E6; 6.92E5 ± 3.15E5; 3.17 ± 0.04). Furthermore following the reference path, this study did not find any scientific approval to justify the index to each dataset. Moreover new combination of the features considered to examine the possibility of defining a similar index. Results failed to give any acceptable credit to such indexing. We have shown that the FLDI is very much biased to the data and is not able to independently and exclusively distinguish healthy data from fatty liver patients. While considering rigorous and robust mathematical analysis of FLDI production, we conclude that high subjectivity of medical data, needs new methodologies for biomedical data fusion and formation of formulas in order to make general quantitative indexing conclusions.
Original language | English |
---|---|
Article number | 102552 |
Number of pages | 10 |
Journal | Biomedical Signal Processing and Control |
Volume | 67 |
Early online date | 16 Mar 2021 |
DOIs | |
Publication status | Published - May 2021 |
Keywords
- Classification of fatty liver
- Non-alcoholic fatty liver disease
- Screening fatty liver
- Tissue features
- Ultrasound liver images
ASJC Scopus subject areas
- Signal Processing
- Biomedical Engineering
- Health Informatics