Purpose: To evaluate the post-radiation lesions of the bone marrow with magnetic resonance imaging (MRI) and image analysis in patients with bone metastases undergoing radiation therapy (RT).
Methods: Thirty-five patients with bone metastases were studied from June 2008 to December 2010. All patients had osseous metastases from various primary malignancies and underwent palliative RT. MRI was performed in a Philips Gyroscan Intera 1T scanner at the beginning of RT and 12-18 days later. T1-TSE, T2-TSE and short tau inversion recovery (STIR) sequences were used. All images obtained were evaluated for early post-radiation lesions. Additionally, 1st and 2nd order textural features were extracted from these images and were introduced into a probabilistic neural network (PNN) classifier in order to create an automated classification system for those lesions.
Results: Changes of signal intensity in T1-TSE, T2-TSE and STIR sequences were evaluated for the presence of edema, fatty conversion of the bone marrow or areas of hemorrhage within the limits of the irradiated area. The automated classification system showed positive results in correctly discriminating the post-radiation lesions that MRI revealed. The overall classification accuracy for discriminating between pre-radiation and post-radiation lesions was 93.2%. Furthermore, the overall classification accuracy for discriminating between post-radiation lesions was 86.67%.
Conclusion: It seems that MRI can evaluate the degree of early therapy-induced bone marrow lesions observed during the first 18 days from the beginning of RT. The proposed neural network-based classification system might be used as an assisting tool for the characterization of these lesions.