When patients are diagnosed with gastric cancer, knowing their microsatellite instability (MSI) status is vital for choosing the right treatment. Currently, identifying this status often happens after surgery. However, researchers analyzed data from over 2,400 patients to see if radiomics—a method of extracting large amounts of data from medical images—could predict this status earlier.
The study found that these imaging models have moderate accuracy in predicting MSI before surgery. While the technology shows promise, it did not perform better than standard logistic regression methods. However, combining these image features with clinical information improved the specificity of the results, which helps doctors better select the right patients for specific therapies.
Because there were differences between how the models performed during training and during validation, researchers note that more rigorous testing is still needed. While radiomics is a helpful tool for prediction, it is not meant to replace the careful judgment of a clinical team.
Common questions
What is radiomics and how does it help with cancer?
Radiomics is a way of pulling large amounts of data from medical images, like CT scans. In this study, it was used to predict microsatellite instability (MSI) in patients with gastric cancer. It helps doctors identify specific markers before surgery, which can help them choose the best treatment plan for each individual patient.
How accurate is this imaging method for predicting results?
The study showed that radiomics has moderate diagnostic accuracy. In a validation set, it achieved an area under the curve (AUC) of 0.82. While machine learning did not outperform standard logistic regression methods, combining image data with clinical features improved the specificity of the prediction.
Is this method ready to replace current doctor assessments?
No, radiomics is not a replacement for clinical assessment by doctors. The study noted that while these tools provide helpful information for patient selection, the results are still subject to some uncertainty because of differences in how models performed during training versus validation.
