Sacroiliitis discrimination method using sacroiliac joint MR image
Disclosed herein is a method for discriminating sacroiliitis by using sacroiliac joint MR images, implemented by one or more processors of a computing device, the method comprising the steps of: collecting MR images related to the sacroiliac joint and preprocessing the collected MR images to generate training data; training a bone marrow edema discrimination model by using the generated training data; and using the trained bone marrow edema discrimination model to determine the presence of sacroiliitis from MR images related to the patient's sacroiliac joint.
1 . A method for discriminating sacroiliitis by using sacroiliac joint MR images through enhanced computer-based image processing techniques, implemented by one or more processors of a computing device, the method comprising the steps of:
collecting MR images related to a sacroiliac joint of a patient using a medical imaging system;
preprocessing the collected MR images;
extracting a region of interest (ROI) corresponding to a sacroiliac joint from the preprocessed MR images by using a trained object detection model comprising a Faster Region-based Convolutional Neural Network (Faster R-CNN), wherein the model is trained to detect anatomical boundaries of the sacrum and ilium;
resizing the extracted ROI to a predefined size;
augmenting the ROI images labeled as positive-class training data using one or more of rotation, blurring, sharpening, contrast adjustment, or noise addition;
generating training data by stacking at least three consecutive MR slices including the ROI;
training a binary classification model comprising a VGG-19-based neural network using the training data; and
using the trained model to determine the presence or absence of sacroiliitis in new MR images of a patient,
wherein the step of preprocessing comprises a step of normalizing the collected MR images to minimize intensity variations in the MR images due to brightness variability between images of different samples or within slices of the samples,
wherein the step of normalizing MR images comprises:
stacking the MR images into a 3D volume;
dividing the 3D volume into a plurality of grid regions;
applying adaptive histogram equalization to each grid region to normalize brightness variations between slices;
converting the processed volume into 2D slices with enhanced contrast and uniform brightness.
2 . The method of claim 1 , wherein the step of extracting regions of interest comprises extracting feature maps from the normalized MR images through the bounding box creation model, calculating the regions of interest of the sacroiliac joint, adjusting the size of the calculated regions of interest to be the same, and performing object classification to extract the final regions of interest (ROI) for each object.
3 . The method of claim 1 , wherein the step of generating training data comprises augmenting data sixfold for images labeled positive, using methods including blurring, contrast adjustment, noise addition, rotation, and sharpening, followed by extracting the area of interest, with the size of the area of interest increased by a set value up, down, left, and right, whereby changes in joint position due to image rotation can be prevented.
4 . The method of claim 1 , wherein the step of generating training data comprises stacking at least three consecutive slices, including the front and back slices based on a certain slice, then generating training data based on the stacked slices to consider the front and rear slices rather than learning with only one slice when determining bone marrow edema.
5 . The method of claim 4 , wherein the step of generating training data is carried out, for both-end slices, by stacking only one slice in front or behind the end slice.
6 . A computing device for enhanced medical image processing, comprising:
a processor comprising at least one of a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), or a tensor processing unit (TPU) configured for medical image analysis; and
a memory optimized for storing 3D medical image volumes,
wherein the processor is configured to perform the method of claim 1 .
7 . A computer program, stored in a non-transitory, computer-readable storage medium and including instructions to cause a computer to perform the enhanced medical image processing comprising the method of claim 1 .