An Overview to Real-Time Automated Liver Segmentation during Laparoscopic Cholecystectomy

Background and aims: During laparoscopic cholecystectomy (the operation of removing gallbladderfrom the patient’s abdomen), liver deformation avoids accurate detection of the liverboundaries. Troubles in detection of vessels and tumor edges of the liver may damage the tissue.Currently, injection of indocyanine green and near-infrared fluorescence imaging technique isused to detect the liver boundaries as well as its vessels, gallbladder, and biliary structure bychanging their color to green in real-time during laparoscopic cholecystectomy. There are somechallenges in using indocyanine green. It takes time for indocyanine green to flow inside thevessels before liver visualization by the fluorescence camera. Moreover, this technique is not permanentand needs to indocyanine green injection per detection. The study objective is to proposean automated deep learning based technique that segments the liver in real-time.Method: In this paper public dataset CholecSeg۸k is employed for network training, validating,and testing. A private dataset from KPJ Damansara specialist hospital, Malaysia is also used justfor testing. For liver segmentation coding, Segmentation Models which is a public library is utilized.U-Net architecture as network model and SE-ResNet۱۵۲ architecture as backbone specifythe border of the liver by using labeled laparoscopic cholecystectomy images as input of thenetwork.Results: Mean IoU score of ۰.۹۶۱۲۷ and mean F-score of ۰.۹۷۹۸۹ are obtained from the validationset in the first top experiment result.Conclusion: Experiments exhibit desirable result of liver segmentation in the CholecSeg۸k dataset.During laparoscopic cholecystectomy, conventional real-time liver segmentation techniquebased on indocyanine green injection and near-infrared fluorescence imaging can be replacedby an automated technique. In the future, improvement of the result for the private dataset willbe investigated. In addition, a robotic arm that handles a laparoscopic camera will be developedwhich will work based of the automated liver segmentation technique proposed here and surgicalinstrument detection.