MRI Radiomics Improves Rectal Cancer Risk Stratification

Artificial intelligence may soon play a larger role in assessing the risk posed by rectal cancer. A new study suggests that analyzing MRI scans with AI-derived radiomics could help clinicians more accurately identify high-risk tumor deposits before surgery, potentially leading to more personalized treatment plans.

A New Approach to Assessing Risk

Researchers conducted a retrospective analysis of 729 patients with rectal cancer treated between 2018 and 2024. The study focused on 376 patients whose data was used to develop and validate models based on radiomics – the extraction of quantitative features from medical images.

Did You Know? Researchers categorized patients into three groups based on the number of tumor deposits: no deposits, one to two deposits, and three or more deposits.

The goal was to determine if these radiomics features could accurately predict the burden of tumor deposits (TD), which is a key factor in determining how the cancer will progress and respond to treatment. Models were built using machine learning, specifically the XGBoost algorithm, and evaluated using metrics like area under the curve (AUC), accuracy, precision, recall, and F1 score.

AI Outperforms Traditional Assessment

A “fusion” radiomics model, combining data from the primary tumor and the largest mesorectal nodule, proved to be the most effective. It achieved an AUC of 0.873 in the test set and 0.858 in the validation cohort, with accuracy approaching 80%. Significantly, this model outperformed two experienced radiologists, whose accuracy ranged from 0.589 to 0.676.

Other models, focusing solely on the tumor or the nodule, also demonstrated strong predictive capabilities, with AUCs exceeding 0.84. Researchers believe that combining information from both the tumor and the largest mesorectal nodule provides a more complete picture of the disease.

Expert Insight: The ability of AI to quantitatively assess tumor deposits represents a potential shift towards more objective and precise risk stratification in rectal cancer. This could move the field closer to tailoring treatment strategies to the individual characteristics of each patient’s disease.

What This Could Mean for Treatment

Tumor deposits are known to be a negative prognostic indicator in rectal cancer, influencing both staging and treatment decisions. However, identifying these deposits accurately using standard imaging can be challenging. The study suggests that MRI-based radiomics could provide an objective tool to help clinicians assess risk before surgery.

This improved risk assessment could lead to more personalized treatment approaches, such as intensified neoadjuvant therapy (treatment before surgery) or closer surveillance following surgery. However, the authors emphasize that the study’s retrospective nature and limited validation mean further research is needed.

Frequently Asked Questions

What are tumor deposits?

Tumour deposits are recognised as an adverse prognostic marker in rectal cancer, influencing staging and therapeutic decisions.

How did the AI models perform compared to radiologists?

The fusion radiomics model outperformed two experienced radiologists, whose accuracy ranged from 0.589 to 0.676.

What type of machine learning algorithm was used?

The models were trained using the XGBoost algorithm.

What role might AI play in the future of rectal cancer treatment?