Radiomics approach with deep learning for predicting T4 obstructive colorectal cancer using CT image

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Obstructive colorectal cancer (OCC) with T4 staging is associated with high mortality.
• Accurate preoperative differentiation between T4 and non-T4 (NT4) stages is critical for treatment planning, especially in emergency settings.

Purpose of the Study:

• To introduce and evaluate three models: radiomics, deep learning, and a combined deep learning-radiomics approach for identifying T4 OCC.
• To assess the contribution of the peritumoral region in T4 OCC detection.

Main Methods:

• A dataset of 164 pathologically confirmed OCC patients' CT images was analyzed.
• Radiomics and deep learning features were extracted and visualized.
• A merged model combining radiomic and deep learning features was developed and evaluated using the area under the receiver operating characteristic curve (AUC).

Main Results:

• The deep learning model achieved an AUC of 0.936.
• The merged deep learning-radiomics model reached an AUC of 0.947, further improving to 0.950 with the addition of clinical features.
• The combined model significantly outperformed individual radiomics and deep learning models.

Conclusions:

• The merged deep learning-radiomics model demonstrates superior performance in classifying T4 OCC compared to standalone models.
• Incorporating the peritumoral region enhances the predictive performance of both radiomics and deep learning models for T4 OCC detection.