# Construction of a prognostic survival model for colorectal cancer patients using CT image texture analysis: a prospective cohort study

## TL;DR

This study builds a survival prediction model for colorectal cancer patients using CT image texture analysis, aiming to improve long-term survival forecasts beyond traditional staging methods.

## Contribution

The novel contribution is the integration of CT image texture features with clinical data to create a more accurate survival prediction model for colorectal cancer.

## Key findings

- Texture features like Teta1, Teta4, and GrSkewness were found to correlate with patient survival times.
- A nomogram was developed to estimate survival and guide treatment decisions.
- Combining radiomics with TNM staging improved the model's predictive accuracy.

## Abstract

Current prognostic indicators for colorectal cancer are limited to pathological staging, which offer only modest predictive value. This study aims to develop a prognostic prediction model for colorectal cancer patients based on texture analysis (TA), with the goal of forecasting long-term survival outcomes.

A total of 236 patients underwent abdominal CT scanning, including both unenhanced and contrast-enhanced CT. Using MaZda software, regions of interest (ROIs) were identified, and texture features were extracted. These texture features were combined with pathological staging data, and statistical analyses were performed using Cox regression, Lasso regression, nomograms, forest plots, receiver operating characteristic (ROC) curve analysis, and survival analysis (Kaplan-Meier curves), and carry out the validation work of the external validation set.

Observation points were established at 1, 3 and 5 years. A correlation analysis was conducted using patient demographic data, tumor markers, pathological staging, and more than 300 variables derived from the texture analysis. The analysis revealed correlations between texture features (such as Teta1, Teta4, WavEnLL_s-2, GrSkewness, and Horzl_RLNonUni) and survival time. Nomograms were created to provide a rough estimation of patient survival, which could assist in decision-making for subsequent treatment plans. Using Lasso regression combined with the nomogram for dimensionality reduction, we were able to intuitively assess the predicted five-year survival time for patients in the perioperative period.

Radiomics analysis of colorectal cancer, when combined with traditional TNM staging, can aid in the construction of a survival prediction model. This model may offer novel insights for predicting long-term survival and provide a reference for the development of individualized treatment strategies.