Assessing the volume of the liver before surgery, predicting the volume of liver remaining after surgery, detecting primary or secondary lesions in the liver parenchyma are common applications that require optimal detection of liver contours, and therefore liver segmentation.

Several manual and laborious, semi-automatic and even automatic techniques exist.

However, severe pathology deforming the contours of the liver (multi-metastatic livers...), the hepatic environment of similar density to the liver or lesions, the CT examination technique are all variables that make it difficult to detect the contours. Current techniques, even automatic ones, are limited in this type of case (not rare) and most often require readjustments that make automatisation lose its value.

All these criteria of segmentation difficulties are gathered in the livers of hepatorenal polycystosis, which therefore constitute an adapted study model for the development of an automatic segmentation tool.

To obtain an automatic segmentation of any lesional liver, by exceeding the criteria of difficulty considered, investigators have developed a convolutional neural network (artificial intelligence - deep learning) useful for clinical practice.

Inclusion Criteria:

• Patients ≥ 18 years old
• Patients with hepato-renal polycystosis, with or without surgery
• Patients with at least one abdominal-pelvic CT scan without injection or with injection between January 1, 2016 and August 2018
• Patients with good quality and available images

Exclusion Criteria:

• Patients with no CT scan images available
• Patients with bad quality of CT scan images

Patients with hepato-renal polycystosis, with or without surgery