Artificial Intelligence-Driven Patient Selection for Preoperative Portal Vein Embolization for Patients with Colorectal Cancer Liver Metastases

Tom N Kuhn ¹, William D Engelhardt ², Vinzent H Kahl ³

¹Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, Connecticut; Department of Diagnostic and Interventional Radiology, University Düsseldorf, Medical Faculty, Düsseldorf, Germany.
²Department of Biomedical Engineering, James McKlevey School of Engineering, Washington University, St. Louis, Missouri.
³Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, Connecticut; Department of Radiology, Charité-Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin, Humboldt-Universität, and Berlin Institute of Health, Berlin, Germany.
⁴Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, Connecticut.
⁵Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, Connecticut; Department of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology, Rostock University Medical Center, Rostock, Germany.
⁶Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, Connecticut; Department of Biomedical Engineering, Yale University, New Haven, Connecticut; Department of Urology, Yale School of Medicine, New Haven, Connecticut.
⁷Interventional Radiology Service, Department of Radiology, Memorial Sloan Kettering Cancer Center, New York, New York.
⁸Department of Clinical Sciences, Florida State University College of Medicine, Tallahassee, Florida.
⁹Hepato-Biliary-Pancreatic Surgery Division, Department of Surgery, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan.
¹⁰Department of Interventional Radiology, The University of Texas MD Anderson Cancer Center, Houston, Texas.
¹¹Department of Surgical Oncology, The University of Texas MD Anderson Cancer Center, Houston, Texas.
¹²Department of Diagnostic and Interventional Radiology, University Düsseldorf, Medical Faculty, Düsseldorf, Germany.
¹³Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, Connecticut; Department of Biomedical Engineering, Yale University, New Haven, Connecticut; Department of Internal Medicine, Section of Digestive Diseases, Yale School of Medicine, New Haven, Connecticut.
¹⁴Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, Connecticut; Department of Internal Medicine, Section of Medical Oncology, Yale School of Medicine, New Haven, Connecticut; Department of Surgery, Section of Surgical Oncology, Yale School of Medicine, New Haven, Connecticut.

⁰Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, Connecticut; Department of Diagnostic and Interventional Radiology, University Düsseldorf, Medical Faculty, Düsseldorf, Germany.

Journal of Vascular and Interventional Radiology : Jvir +

|

December 5, 2024

View abstract on PubMed

Summary

Machine learning models accurately predict outcomes after portal vein embolization (PVE) for colorectal cancer (CRC) patients, aiding surgical decisions. These models use radiomics and lab data to forecast future liver remnant (FLR%) and kinetic growth rate (KGR%).

Area Of Science

Hepatobiliary surgery
Medical imaging
Machine learning

Background

Colorectal cancer (CRC) liver metastases often require hepatic resection.
Portal vein embolization (PVE) is used to increase future liver remnant (FLR) volume before surgery.
Predicting PVE outcomes is crucial for selecting appropriate surgical candidates.

Purpose Of The Study

To develop machine learning algorithms for predicting post-portal vein embolization (PVE) outcomes in metastatic colorectal cancer (CRC) patients.
To improve the selection of patients eligible for hepatic resection by accurately forecasting liver remnant function and growth.
To integrate radiomic features and laboratory values into predictive models.

Main Methods

A multicenter retrospective study of 200 patients with CRC liver metastases undergoing PVE.
Collected radiomic features and laboratory values; calculated patient-specific liver shape eigenvalues.
Trained three machine learning models to predict total liver volume (TLV), sufficient future liver remnant (FLR%), and kinetic growth rate (KGR%), with validation on internal and external test sets.

Main Results

Machine learning models demonstrated high prediction accuracy and AUC for sufficient FLR% (internal: 85.81%, AUC 0.91; external: 79.66%, AUC 0.88) and KGR% (internal: 87.44%, AUC 0.66; external: 72.06%, AUC 0.69).
Total liver volume (TLV) prediction showed discrepancy rates of 12.56% internally and 13.57% externally.
Performance metrics were consistent across internal and external validation sets, indicating model robustness.

Conclusions

Machine learning models integrating radiomics and laboratory data can effectively predict FLR%, KGR%, and TLV.
These predictive capabilities can serve as valuable metrics for assessing the success of PVE.
The developed models may enhance the selection process for hepatic resection in CRC patients with liver metastases.