Skip to main content
SSO 2024 Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Presentation Icons
Recorded for On-Demand Viewing
Recorded for On-Demand Viewing
Back

HPB

HPB Parallel Session 2

57: Predicting Local Recurrence of Colorectal Liver Metastasis After Hepatectomy Using Artificial Intelligence

Saturday, March 23, 2024
9:12am – 9:22am ET
Location: C105-106
Disclosure(s):
Hardik Y. Patel, D.O.: No financial relationships to disclose
Introduction: Approximately 20% of patients diagnosed with colorectal cancer have hepatic metastases, and another 50% will develop metastases in their lifetime. For select patients with Colorectal Liver Metastasis (CRLM), hepatic resection is the treatment of choice; however, recurrence rates are high. Current prognostic scores, such as the Fong Score, have poor predictive performance with regards to recurrence after surgery and were designed before the era of neoadjuvant treatment. In this study we leverage Artificial Intelligence (AI) to predict local recurrence after hepatectomy for CRLM.


Methods: A retrospective analysis of a publicly available dataset including patients who have undergone hepatectomy for CRLM with available two-year follow up data, preoperative segmented Computed Tomography (CT) scans, and relevant clinical data were included. A Machine Learning (ML) model was trained using calculated radiomic and clinical features with 10-fold cross-validation which was then fused with a Deep Learning (DL) model to predict 2-year local recurrence after hepatectomy. The primary outcomes were model accuracy and Area Under the Curve of the Receiver Operating Curve (AUC-ROC) with a secondary outcome of predicted survival.


Results: A total of 168 patients with 453 CRLM were included, with an average of 2.7 metastatic lesions per patient (SD: 0.2) and 33% hepatic recurrence rate after hepatectomy within 2 years. A clinical model consisting of 13 pre-operative clinical features was trained yielding an accuracy of 66% and AUC of 0.7. A radiomic model consisting of 108 calculated radiomic features reduced to 3 principal components with an explained variance of 94% was trained yielding an accuracy of 61% with AUC of 0.64. A DL model utilizing a DenseNet architecture was trained on preoperative CT imaging yielding an accuracy of 70% and AUC of 0.65. A fusion Naïve Gaussian Bayes model was then created combining the clinical, radiomic, and DL features to attain an accuracy of 74% and an AUC of 0.87. Kaplan-Meier analysis of predicted class shows an improved median overall survival (mOS) of 103 months versus 47 months (P< 0.001) in patients not predicted to have a 2-year recurrence.  


Conclusions: The application of AI in predicting local recurrence of CRLM can aid surgeons in making complex decisions for their patients. Further studies with larger datasets incorporating genomic data are needed to create a clinically robust model.

Learning Objectives:

  • Upon completion, participant will able to describe the characteristics of a fusion model.
  • Upon completion, participant will able to understand the application of computer vision in the care of patients with colorectal liver metastasis.
  • Upon completion, participant will able to participate in discussions involving deep learning in surgery.