Radiological staging clinical decision support model for rectal cancer lymph node metastasis detection on MRI.

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Accurate staging of lymph node metastasis (LNM) is critical for personalized rectal cancer treatment.
• Malignant lymph nodes are a primary site for rectal cancer metastasis, often treated with neoadjuvant chemoradiotherapy (CRT).
• Current radiological staging relies on subjective criteria, leading to variable radiologist performance (73% sensitivity, 74% specificity).

Purpose of the Study:

• To develop an automated, end-to-end artificial intelligence (AI) model for pre-operative radiological staging of rectal cancer lymph node metastasis.
• To integrate automatic lymph node detection on MRI with patient-level staging using multiple instance learning.
• To evaluate the AI model's performance against expert radiologists.

Main Methods:

• Utilized an in-house dataset of 458 patients with pre-operative MRI scans and pathological TNM staging.
• Employed nnU-Net for accurate lymph node detection on MRI.
• Developed a multiple instance learning framework for patient-level LNM staging, comparing various 3D feature encoders.

Main Results:

• Achieved state-of-the-art performance with 0.828 AUC, 86.6% sensitivity, and 72.4% specificity (cross-validated).
• The AI staging model demonstrated superior performance compared to three expert radiologists in predicting post-operative pathology.
• The AI model achieved a 9% higher F1 score than expert radiologists in clinical validation.

Conclusions:

• An end-to-end AI model significantly enhances pre-operative staging of rectal cancer lymph node metastasis.
• The AI model's performance surpasses that of expert radiologists, offering potential for improved clinical decision-making.
• This AI approach shows promise in improving patient outcomes through more accurate staging.