Embedded Files
To validate the noise correction method used in our protocol, we visually assessed T2 anatomical, RSI images at b-values of 0 and 3000 s/mm2 , and noise maps before and after noise correction. ADC maps were generated at the slice depicting both the ovary and the lesion, and regions of Interest (ROIs) were drawn on both the patient’s normal ovary and their lesion. These ROIs generated ADC values that were recorded both before and after noise correction. The metric used for the comparison was percent error, which is calculated by the following equation:
Quantitative Analysis of ADC Values
Quantitative Analysis of ADC Values
Literature Benign Tumor ADC Value: https://www.sciencedirect.com/science/article/pii/S0378603X1400062X?via%3Dihub
Literature Borderline and Malignant Tumor ADC Values: https://link.springer.com/article/10.1007/s11604-020-00960-2
The table above displays a quantitative analysis between the expected, pre-noise correction, and post-noise correction ADC values for the benign, borderline and malignant ovarian tumors. Expected ADC values were found in primary research papers for each tumor type and ADC values were evaluated on each ADC Map using MATLAB. For each tumor, an ADC map produced after noise correction was applied to the DWI images decreased the error to the expected ADC value. The error decreased from 55% to 15% in the benign lesion, 23% to 3% in the borderline lesion, and 17% to 9% in the malignant lesion. The images for each case are linked below for reference.
Final Thoughts
Final Thoughts
Noise-corrected images seem to significantly decrease signal at higher b-values. We also observed that the current noise correction method filters out the body noise of RSI images, indicating that noise is spatially dependent due to the variance of color in the noise maps. We also noted through our analyzed cases that the ADC maps without filtering results in more stable ADC values. However, a key observation made in our results is that the ADC values produced with noise-corrected images are closer to the ADC values expected in literature.
The developed noise correction algorithm not only correctly filters out noise, allowing for physicians to make more accurate qualitative observations about a patient scan, but also produces more reliable ADC values. In a field, such as radiology, where much of clinical decision making is driven by qualitative results in comparison to quantitative results, this noise correction algorithm serves to create more trust in quantitative values while also providing a lower potential rate of false negatives and positives if used.
Page Author: Rahul Sehgal
Page updated
Report abuse