International Congress of Radiology 2023, Al-Ghardaqah, Egypt, 15 - 17 March 2023, pp.98
Prediction of osteoporosis from magnetic resonance imaging (MRI) of the sacrum with using texture
analysis-based machine learning algorithms.
Patients who underwent sacroiliac MRI between 2018 and 2021 and had bone mineral densitometry
(BMD) measurement were included the study. Axial slices of sacrum at the S2 level from T1 sequences
were saved as DICOM format. Then, the images were loaded into the texture analysis software MaZda
(version 4.6). ROIs were placed in the sacrum without including the and surrounding fat tissue with the
consensus of two radiologists. From each patient, 312 features were extracted by texture analysis. The
results were transferred to Orange Data Mining software (Version 3.33.0), and classification was
performed. At this stage, the classification process was performed using the 5-fold cross-validation
The mean age of 5 male and 101 female patients was 54.3 years. Forty-three patients were diagnosed
with osteopenia or osteoporosis by BMD results. The most successful algorithms in the area under the
curve (AUC) result were Neural Network, Gradient Boosting, and Naive Bayes, and their ratios were
0.72,0.70,0.69, respectively. Classification accuracy rates are 0.69,0.68,0.67, and specificity rates are
0.68,0.65,0.68, respectively. The neural network algorithm, successfully predicted 45 out of 63 patients
who has normal BMD results.
Osteoporosis is an important pathology that increases the risk of bone fractures. Although more studies
are needed, it is promising to be able to predict osteoporosis with Texture analysis-based machine
learning algorithms from MRI images with no use of ionizing radiation.