/MPBD-LSTM

MPBD-LSTM: A Predictive Model For Colorectal Liver Metastases Using Time Series Multi-phase Contrast-Enhanced CT Scans

MPBD-LSTM: A Predictive Model For Colorectal Liver Metastases Using Time Series Multi-phase Contrast-Enhanced CT Scans

This is the official repo of the article "MPBD-LSTM: A Predictive Model For Colorectal Liver Metastases Using Time Series Multi-phase Contrast-Enhanced CT Scans" (MICCAI 2023). Code will be uploaded shortly.

Dataset:

Our dataset consists of 269 patients in total, with a positive rate of 25.7%. Aside from the plain scan phase, it has 2 additional contrast-enhanced phase: the portal venous phase (V) and the arterial phase. It's collected based on the following rules:

  • No tumor appears on the CT scans. That means patients have not been diagnosed as CRLM when they took the scans.
  • Patients were previously diagnosed with colorectal cancer TNM stage I to stage III, and recovered from colorectal radical surgery.
  • Patients have two or more times of CECT scans.
  • We already determined whether or not the patients had liver metastases within 2 years after the surgery, and manually labeled the dataset based on this.
  • No potential focal infection in the liver before the colorectal radical surgery.
  • No metastases in other organs before the liver metastases.
  • No other malignant tumors.

For more details, please refer to the article.
We are further doing more experiments for the task, will upload the code once it's finished.

Please send email to the author at xli34@nd.edu for the link and password to download the dataset.