Background: Assessment of hepatic steatosis (HS) before transplantation requires the pathologist to read a graft biopsy. A simple method based on the evaluation of images from tissue samples with a smartphone could expedite and facilitate the liver selection. This study aims to assess the degree of HS by analysing photographic images from liver needle biopsy samples. Methods: Thirty-three biopsy-images were acquired with a smartphone. Image processing was carried out using ImageJ: background subtraction, conversion to HSB colour space, segmentation of the biopsy area, and evaluation of statistical features of Hue, Saturation, Brightness, Red, Green, and Blue channels on the biopsy area. After feature extraction, correlations were made with gold standard HS percentage assessed at two levels (frozen-section vs glass-slide). Sensitivity, specificity, and accuracy were calculated for each feature. Results: Correlations were found for H, S, R. The sensitivity, specificity, and accuracy of the final classifier based on the K* algorithm were 94%, 92%, 94%. Limitations: Accuracy assessment was performed considering macrovesicular steatosis on specimens with mostly < 30% HS. Conclusions: The steatosis assessment based on needle biopsy images, proved to be an effective and promising method. Deep learning approaches could also be experimented with a larger set of images.
Assessment of hepatic steatosis based on needle biopsy images from deceased donor livers
Della Mea V.;Terrosu G.;Pravisani R.;Di Loreto C.;Risaliti A.;Baccarani U.
2021-01-01
Abstract
Background: Assessment of hepatic steatosis (HS) before transplantation requires the pathologist to read a graft biopsy. A simple method based on the evaluation of images from tissue samples with a smartphone could expedite and facilitate the liver selection. This study aims to assess the degree of HS by analysing photographic images from liver needle biopsy samples. Methods: Thirty-three biopsy-images were acquired with a smartphone. Image processing was carried out using ImageJ: background subtraction, conversion to HSB colour space, segmentation of the biopsy area, and evaluation of statistical features of Hue, Saturation, Brightness, Red, Green, and Blue channels on the biopsy area. After feature extraction, correlations were made with gold standard HS percentage assessed at two levels (frozen-section vs glass-slide). Sensitivity, specificity, and accuracy were calculated for each feature. Results: Correlations were found for H, S, R. The sensitivity, specificity, and accuracy of the final classifier based on the K* algorithm were 94%, 92%, 94%. Limitations: Accuracy assessment was performed considering macrovesicular steatosis on specimens with mostly < 30% HS. Conclusions: The steatosis assessment based on needle biopsy images, proved to be an effective and promising method. Deep learning approaches could also be experimented with a larger set of images.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
