Inconsistent Image Prediction
This section is dedicated to helping you identify and resolve common problems that may affect the consistency and reliability of prediction scores.
Probable Causes and Solutions
Encountering issues with the accuracy of qXR image prediction scores? It can be due to the following reasons.
Low Resolution: The image lacks detail, making accurate analysis difficult.
Solution: Ensure the scan pixel size is at least 1400*1400 pixels to achieve optimal clarity and detail necessary for accurate predictions.
Insufficient Bit Depth: The image appears flat, lacking the depth required for a detailed analysis.
Solution: Confirm that the scan's bit depth is a minimum of 8 bits. This depth ensures the image has enough information for a thorough analysis.
Clipped Lung Anatomy: Missing parts of the lung anatomy in the image, leading to incomplete analysis and inaccurate predictions.
Solution: Ensure the entire lung anatomy is visible in the scan. Avoid using images with clipped anatomy as they are not suitable for analysis.
Incorrect Patient Position: The lung anatomy may appear distorted or partially obscured due to the patient's rotation or inadequate inspiration during the scan.
Solution: The patient should not rotate and must fully inspire at the time of scanning. This positioning captures the lung anatomy in the best position for accurate analysis.
Suboptimal Exposure: Critical details of the lung anatomy are missed either because the image is too dark or too light. Solution: Adjust exposure to optimal levels to capture all necessary details of the lung anatomy for reliable prediction scores.
Improper Image Composition: The presence of non-chest views or inclusion of other body parts in the scan can compromise prediction accuracy.\
Solution: Use only chest view images for analysis. Avoid grainy images and photos of scans to ensure the accuracy of prediction scores.
By systematically addressing each of these probable causes with the suggested solutions, users can significantly improve the accuracy of qXR image predictions.
Last updated
