Manuela Cesaretti
Hospital Beaujon, France
Title: Liver-donor steatosis assessment from smartphone images acquired in the operatory room.
Biography
Biography: Manuela Cesaretti
Abstract
Fast and accurate graft hepatic steatosis (HS) assessment is of primary importance for lowering liver dysfunction risks after transplantation. Despite diagnosis being recognized as challenging in the clinical literature, few efforts have been invested to develop computer-assisted solutions for HS assessment. The objective of this study is to investigate the automatic learning-based analysis of liver texture from RGB images acquired in the operating room (OR), with the goal of classifying liver grafts rejected due to high HS level. Forty RGB images of forty different donors were analyzed. Twenty images refer to accepted livers and twenty to livers that were discarded due to high HS value. The ground-truth HS diagnosis associated to each image was obtained with histopathological analysis. The images were captured with an RGB (12-megapixel) smartphone camera in the OR. Intensity-based features ( S tat 1), histogram of local binary pattern ( HLBP) and features extracted from blood tests ( B l o) were investigated. Semi-supervised multiple instance learning was exploited to perform image classification (Fig 2). With the best performing feature (HLBP + S tat 1 + B l o), the overall classification accuracy was 0.88. The achieved recall in classifying discarded grafts was 0.95. The results suggest that the proposed method is a promising strategy to develop a fully automatic solution to assist surgeons in HS assessment inside the OR.