| Autor: |
DICKSON, A. JOEL, LINSELY, J. ARUL, NINETA, R. J. ALICE |
| Předmět: |
|
| Zdroj: |
Oxidation Communications; 2023, Vol. 46 Issue 2, p328-336, 9p |
| Abstrakt: |
Liver cancer is a major killer worldwide, and it is also among the most frequent types of cancer. Methods for automatically segmenting and classifying liver tumours are crucial for aiding physicians in making a diagnosis. Powerful categorisation algorithms with the potential to alter many practical applications are becoming available as Artificial intelligence (AI) develops. Liver tumour classification is challenging because of noise, non-homogeneity, and significant appearance variations in cancer tissue. In this study, we provide a novel approach to automatically segmenting and classifying liver tumours. There are three primary components to the suggested structure. To begin, the picture contrast is improved with the help of a preprocessing unit. In addition, we suggest using a Masked Recurrent Convolutional Neural Network (RCNN) to segment the liver. The final characteristic maps of the liver for abnormalities identification are generated by performing a pixel-wise classification unit. Dice similarity coefficients of 96 and 98% are achieved, respectively, for liver segmentation and lesion identification, when suggested models are applied to the demanding MICCAI'2027 liver tumour segmentation (LITS) database. The efficiency of the proposed framework is shown by comparisons with similar strategies for tumour segmentations. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
