# Baseline MRI habitat imaging for predicting treatment response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer

## TL;DR

This study shows that baseline MRI habitat imaging can help predict how well patients with rectal cancer will respond to a specific treatment before surgery.

## Contribution

A new clinical-habitat model combining MRI habitat imaging and clinical data improves prediction of treatment response in rectal cancer.

## Key findings

- The ModelClinic+Habitat achieved an AUC of 0.710 in the validation set, outperforming other models.
- Habitat imaging provides clinically interpretable spatial heterogeneity information for treatment prediction.
- The model may support personalized treatment decisions for locally advanced rectal cancer.

## Abstract

This study was to assess whether baseline magnetic resonance habitat imaging can predict the efficacy of neoadjuvant chemoradiotherapy (nCRT) in patients with locally advanced rectal cancer (LARC).

This retrospective study analyzed data from 181 patients with locally advanced rectal cancer, including 60 who exhibited a good treatment response. The cohort was randomly divided into a training set (127 patients, 42 with good response) and a validation set (54 patients, 18 with good response). Five models were developed: ModelClinic, ModelRadiomics, ModelHabitat, ModelClinic+Radiomics, and ModelClinic+Habitat. Model performance was assessed using the area under the receiver operating characteristic (ROC) curve (AUC) for both training and validation sets.

The AUC values for predicting the efficacy of LARC neoadjuvant therapy were as follows: in the training set, ModelClinic achieved 0.788, ModelRadiomics 0.827, ModelHabitat 0.815, ModelClinic+Radiomics 0.938, and ModelClinic+Habitat 0.896; in the test set, the corresponding AUCs were 0.656, 0.619, 0.636, 0.532, and 0.710, respectively. Decision curve analysis demonstrated that the clinical combined habitat model (ModelClinic+Habitat) provided higher net benefits than other models within a threshold probability range of 20% to 80%.

The habitat model we developed, which integrates first-order and clinical features, demonstrates potential for predicting the efficacy of nCRT clinically interpretable spatial heterogeneity information. This model may aid in personalized treatment decision-making for LARC.