Week 1 Shadowing in Radiology Reading Room
I am Xinzi He, work with Prof. Mert Sabuncu at Cornell Tech. I didn't meet my clinic mentor this week but I worked with Dr Prince and did shadowing in the radiology reading room to see how radiologists use various sequences to do segmentation, radiometric measurement, and diagnosis of ADPKD (autosomal dominant polycystic kidney disease) patients. In abdominal MRI for instance, there are four usually used sequences: Axial T1, Coronal T1, Axial T2, Coronal T2 and Axial SSFP. The most popular way to investigate the process is so called longitudinal study that is measuring the volume change of kidneys over time. Due to the motion of the lungs, abdominal MRI scans usually are anisotropic. For example, an Axial scan means that physical resolutions over the axial plane is higher than the thickness. This leads to an issue that in order to accurately measure the volume of kidneys of one patient we have to aggregate segmentation results from both Axial and coronal scans. This means we need to do segmentation multiple times for one patient.
However, measurement via manual segmentation is tedious, laborious and objective. Therefore, to improve efficiency and accuracy, Dr. Prince seeks to use artificial intelligence to automatically segment them. However, training a model which can accurately segment all organs in abdominal images: kidneys, spleen, liver, blade and stomach requires tremendous labeled scans. Thus, in addition to learning how to diagnose diseases via MRI images, we were responsible for Multi-organ, multi-sequence MRI image segmentation using a tool named ITK-SNAP. ITK-SNAP is a segmentation tool designed for 3D medical image manual segmentation. Unlike the segmentation on normal patients, ADPKD multi-organ segmentation requires lots of expertise besides the anatomy of kidneys because kidneys are usually swelled and occupied some other organs regions.
In the next week, I wanna learn how to train a model for automatic segmentation and go to neuroICU.
Comments
Post a Comment