Prediction of Bone Marrow Metastases Using Computed Tomography (CT) Radiomics in Patients with Gastric Cancer: Uncovering Invisible Metastases.

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Gastric cancer commonly metastasizes to bone.
• Distinguishing bone metastases not visible on computed tomography (CT) from unaffected bone is challenging.
• Pathological confirmation is the current reference standard for bone metastasis.

Purpose of the Study:

• To investigate if radiomics of CT image data can differentiate bone metastases from unaffected bone in gastric cancer patients.
• To assess the predictive performance of machine learning models using CT-derived features.
• To evaluate radiomics as a potential imaging biomarker for bone marrow metastasis.

Main Methods:

• Retrospective study including 96 gastric cancer patients with pathologically confirmed bone metastasis.
• Feature sets included CT attenuation values, radiomic features, and combined features from the region of interest.
• Five machine learning models were developed and evaluated, with the best performing model validated in an external cohort.

Main Results:

• A Random Forest classifier using combined radiomics and attenuation data achieved the highest predictive performance (AUC, 0.96).
• The model demonstrated excellent accuracy (AUC, 0.93) even in the pathology-positive, CT-negative subgroup.
• Internal and external validation confirmed the model's consistent and excellent predictive accuracy.

Conclusions:

• Radiomic features from CT images are effective imaging biomarkers for predicting bone marrow metastasis in gastric cancer.
• This approach shows potential for clinical utility in diagnosing and predicting bone marrow metastasis via routine abdominopelvic CT evaluation.
• Radiomics offers a non-invasive method to enhance the detection of bone metastases in gastric cancer follow-up.