Automatic Generation of Liver Virtual Models with Artificial Intelligence: Application to Liver Resection Complexity...

Area of Science:

• Hepatobiliary surgery
• Medical imaging
• Artificial intelligence in medicine

Background:

• Liver resection (LR) is a primary treatment for liver cancer, but high mortality/morbidity rates persist.
• Current methods for classifying liver resection complexity (LRC) do not account for 3D anatomical complexity induced by disease.

Purpose of the Study:

• To develop and validate an AI-driven tool for predicting intraoperative liver resection complexity (LRC) using only preoperative CT scans.
• To create a novel anatomical framework for assessing surgical complexity based on the Hepatic Central Zone (HCZ).

Main Methods:

• 3D organ, tumor, and vessel models were generated using deep learning on patient CT scans.
• An automated pipeline was developed to define the HCZ and quantify tumor proximity.
• An AI model was trained on 145 HCC patients to predict LRC, comparing its performance against surgeon predictions.

Main Results:

• Accurate 3D reconstructions and HCZ generation (Dice score 82±4.6%) were achieved, even with atypical vasculature.
• The AI model demonstrated superior LRC prediction accuracy (79.4±3.4%) and AUC (85.1±3.2%) compared to surgeons.
• The automated pipeline successfully processed a cohort of 145 HCC patients.

Conclusions:

• An automated digital tool accurately predicts intraoperative LRC from preoperative CT scans.
• This technology offers innovative potential for oncology surgery planning and patient referral.
• The tool can assist in directing patients to specialized medical centers based on predicted surgical complexity.