The Use of Artificial Intelligence for the Prediction of Recurrence After Resection of Colorectal Liver Metastases (AI-RECOLMET)

A Retrospective Observational Study to Use Artificial Intelligence for Prediction of Disease REcurrence of COlorectal Cancer Liver METastasis After Hepatic Resection

Colorectal cancer is the third most common cancer worldwide and the fourth most common cause of cancer-related death. Survival is primarily determined by stage of disease and the presence of metastases. The combination of chemotherapy and liver resection remains the treatment option with the highest survival benefit for patients with liver metastases from colorectal cancer, with surgery still being the only recognized potential curative treatment; surgical locoregional treatment can also be combined with thermal ablation to enhance the possibility of complete liver clearance. Despite significant improvements in prognosis, a large proportion of patients (almost half) will still experience recurrence following treatment. There is a clinical need to identify a priori patients who are different likely to develop disease recurrence after locoregional treatment (liver resection ± thermal ablation) and to respond differently to chemotherapy, in order to refine risk-based allocation of treatments and resources. Widespread digitalization of healthcare generates a large amount of data, and together with today accessible high-performance computing, artificial intelligence technologies can be applied to overcome the current limitations in estimating colorectal cancer liver metastases recurrence and response to locoregional and chemotherapy treatments, thus achieving better treatment allocation than current practice. All radiomic features can also help in training the neural network aimed at detecting liver metastases before they become visually detectable by the radiologist. Therefore, this study aims to evaluate whether a multifactorial machine learning model (including clinical and radiomic) can identify patients with colorectal cancer liver metastases with a high risk of progression after chemotherapy and recurrence after liver resection

Study Overview

• Status: Recruiting
• Conditions: Hepatectomy; Liver Ablation; Liver Resection; Colorectal Liver Metastasis (CRLM)
• Intervention / Treatment: Other: AI-analysis

Detailed Description

ColoRectal Cancer (CRC) is the third most common cancer worldwide and the fourth most common cause of cancer-related deaths. Survival is mainly determined by disease stage and the presence of metastasis. Five-year survival among patients with metastatic CRC is 12-19% versus 90% in patients with localized disease, with ColoRectal cancer Liver Metastases (CRLM) occurring in 30-50% of patients with CRC and being responsible for two-thirds of CRC-related deaths. Recent advances in the treatment of CRLM have increased patient overall survival (OS) from 6 months to 5-year survival rates of 25-40%. The combination between chemotherapy and liver resection remains the therapeutic option with the highest survival benefit for patients with CRLM. With surgery still representing the only acknowledged potential curative treatment, on the other hand, modern chemotherapy, also thanks to the introduction of biological drugs, has led to a better response and survival rates for patients with liver metastases, also contributing to increasing the resectability rate (i.e. the number of patients made operable thanks to cytoreduction following medical therapy).

Up to 30% of patients may be cured if metastases in the liver can be completely removed (the medical term for this is "resection"). For surgery to be considered, an oncological disease must be radically resectable from the technical point of view. At the same time, an adequate amount of normal liver must be left behind after the resection to sustain life. Locoregional therapies, including thermal ablation, chemoembolization, and radiation, are also used to manage CRLM patients as alternatives to conventional curative treatment. In particular, the locoregional treatment of CRLM can benefit from thermal ablation, either radiofrequency-based or microwave-based. Indeed, even if resection remains the locoregional treatment of choice for resectable liver metastases, ablation may offer similar benefits in selected patients, helping to spare healthy liver parenchyma.

Despite these significant improvements in prognosis, a large proportion of patients (nearly half) will anyway experience recurrence following the combination of locoregional treatments. With the improvement of non-surgical therapies, patients who relapse could undergo a non-surgical treatment rather than resection. Therefore, it is of paramount importance to define at best the treatment strategy for patients based on an accurate estimate of prognosis after treatment, by balancing the perioperative risk of morbidity/mortality with the risk of recurrence. Identifying these relapsing patients in advance would be crucial to avoid futile surgery or to allocate them to pharmacological adjuvant (i.e. postoperative) treatments.

Most patients with CRLM undergo preoperative chemotherapy programs with a response probability of around 60% of cases. The healthcare and biological costs of chemotherapy programs are significant. Furthermore, several chemotherapy regimens currently exist, but there is a lack of data indicative of which regimen is effective in which patient. Identifying these responding patients in advance would be extremely determinant to avoid futile chemotherapy treatment without clinical benefit and to guide the choice of allocating them or not to pharmacological neoadjuvant (i.e. preoperative treatments) and the selection of adjuvant regimen. In turn, it is important to distinguish responders from non-responders early to select the most appropriate therapeutic approach.

The clinical characteristics of the patients and the disease, in addition to radiological imaging, are all proved to be indispensable tools that help to evaluate the extent of disease, assess response to treatment, and identify drug toxicities and recurrence. Some studies suggest that texture feature analyses may quantitatively detect liver metastases before they become visually detectable by the radiologist. However, the value of these conventional factors alone in predicting CRLM prognosis is restricted. Multifactoral prognostic scoring systems have been developed in the last years as potential recurrence predictors after liver resection, but these still host limitations in applicability, sensitivity and specificity. Novel accurate prognostic indicators in patients with CRLM are urgently needed. Specifically, there is a clinical need to identify a priori patients who have different probabilities of developing recurrence of the disease after locoregional treatment (liver resection with or without thermal ablation) and different response to chemotherapy treatment, in order to refine a risk-based allocation to treatments and of resources. As known, Artificial Intelligence (AI) is a branch of computer science that aims to simulate human intelligence and behaviors to assist humans in specific tasks. The widespread digitalization of healthcare generates a vast amount of data and, together with accessible high-performance computing, AI technologies can be applied to overcome actual limitations in the estimation of CRLM recurrence and response after locoregional and chemotherapeutic treatments, thus reaching a finest allocation to treatments with respect to the current practice.

Computed tomography (CT) is the imaging of choice for staging and follow-up in most patients with CRLM. However, many factors could influence the recurrence and OS. For example, up to 25% of lesions may go undetected. A meta-analysis reported that contrast-enhanced CT had a sensitivity and specificity of 82% and 84%, respectively, for detecting liver lesions. However, for lesions smaller than 1 cm, the sensitivity drops to 31-38%. MRI is the most accurate modality for detecting CRLM; however, it is typically used as a secondary tool in practice. MRI has the advantage of being able to detect and characterize lesions even smaller than 10 mm as well as no radiation hazard. Hepatocyte-specific contrast agents have a sensitivity of 95% for detecting liver metastasis.

Despite their value visual CT and MRI evaluation fails to assess micro-structural changes like intra-lesional perfusional or necrotic changes that might be early indicators of a high risk for local tumor progression. Also, a bad imaging quality could lead to the inability to recognize all the metastases. Therefore, the interest in the use of AI-based tools to optimize imaging quality and the use of quantitative imaging has been growing with the use of texture analysis or radiomics. Radiomics consists of extracting a large amount of quantitative features from medical images in combination with machine learning (ML) models to predict clinical endpoints. Recently, radiomics have been studied as a prognostic/predictive indicator for clinical outcomes in different tumor types and survival analysis with promising results on CLRM treated with microwave ablation. However, little evidence is available for the liver and for CRLM and some studies suggest that texture feature analyses may quantitatively help the radiologist in modifying clinical approaches.

All the radiomic features can also help in training neural network aimed to detecting liver metastases before they become visually detectable by the radiologist.

Therefore, this study aims to assess whether a multi-factoral (including clinical and radiomics) ML model can identify patients with CRLM with a high risk for progression after chemotherapy and recurrence after liver resection. The modeling phase will take place on S-RACE by D34Health, based on Azure Machine Learning on Azure Cloud, providing useful tools and methods required for modeling the the problem, such as compute instances and modeling tools, like Jupyter Notebooks, coupled with the built-in collaboration and privacy tools.

Furthermore, the added value of clinical data could be used to build clinical and combined models for better outcomes prediction.

The study is configured as an observational, retrospective study, monocentric. Subjects affected by CRLM and operated of liver resection in our hospital environment will be selected according to eligibility criteria.

• Study Type: Observational
• Enrollment (Estimated): 1000

Contacts and Locations

• Study Contact: Name: Francesco De Cobelli, MD; Phone Number: 390226432529; Email: trialcliniciradiologia@hsr.it
• Study Contact Backup: Name: Stephanie Steidler, PhD; Phone Number: 390226432529; Email: trialcliniciradiologia@hsr.it

Study Locations

• Italy
  • Milan, Italy, 20123
    • Recruiting
    • Radiology Department
    • Contact: Francesco De Cobelli, MD; Phone Number: 390226432529; Email: trialcliniciradiologia@hsr.it

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Probability Sample

Study Population

Patients with colorectal liver metastases

Description

Inclusion Criteria:

• Pathologically confirmed diagnosis (at final pathology) of liver metastases from colon or rectal adenocarcinoma
• > 6 months of follow-up
• no other concomitant neoplastic disease

Exclusion Criteria:

• All subjects receiving hepatic resection but not fulfilling the inclusion criteria

Study Plan

How is the study designed?

Number of groups / cohorts

1

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Patients with CRLM treated with liver resection (with or without liver ablation); Patients with colorectal cancer liver metastases receiving liver resection (with or without liver ablation) with or without perioperative (pre- , post- or pre-post-) systemic chemotherapy.	Other: AI-analysis; The study will investigate machine learning models to predict recurrence after liver resection for CRLM

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Development of an ML algorithm predicting which individuals diagnosed with CRLM are most likely to experience early recurrence of disease after liver resection.	The primary endpoints of this clinical study are the sensitivity, specificity, and area under the Receiver Operating Characteristic (AUC-ROC) curve of the machine learning models in predicting oncological outcomes: early recurrence based on clinical and radiological features.	6 months post-intervention

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Development of an ML algorithm predicting which individuals diagnosed with CRLM are most likely to experience early recurrence of disease after liver resection	Sensitivity, specificity, and area under the Receiver Operating Characteristic (AUC-ROC) curve of the machine learning models in predicting time to recurrence, disease free survival (DFS), progression free survival (PFS), recurrence free survival (RFS), overall survival (OS), based on clinical and radiological features	Through study completion, an average of 18 months
Development of a ML algorithm predicting which individuals diagnosed with CRLM are most likely to experience response of disease to neoadjuvant systemic chemotherapy	Sensitivity, specificity, and area under the Receiver Operating Characteristic (AUC-ROC) curve of the machine learning models in predicting PFS and OS based on the pre-treatment contrast-enhanced CT or MRI scan	Through study completion, an average of 18 months

Collaborators and Investigators

• Sponsor: Francesco De Cobelli

Study record dates

Study Major Dates

• Study Start (Actual): February 19, 2025
• Primary Completion (Estimated): February 19, 2026
• Study Completion (Estimated): February 19, 2027

Study Registration Dates

• First Submitted: January 22, 2026
• First Submitted That Met QC Criteria: January 29, 2026
• First Posted (Actual): February 4, 2026

Study Record Updates

• Last Update Posted (Actual): February 4, 2026
• Last Update Submitted That Met QC Criteria: January 29, 2026
• Last Verified: January 1, 2026

More Information

Terms related to this study

• Keywords: Hepatectomy; Liver resection; Artificial Intelligence (AI); Liver ablation; Colorectal liver metastases; Machine learning (ML)
• Other Study ID Numbers: AI-RECOLMET

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No