Radiomics and Machine Learning Can Differentiate Transient Osteoporosis from Avascular Necrosis of the Hip

Abstract

Differentiation between transient osteoporosis (TOH) and avascular necrosis (AVN) of the hip is a longstanding challenge in musculoskeletal radiology. The purpose of this study was to utilize MRI-based radiomics and machine learning (ML) for accurate differentiation between the two entities. A total of 109 hips with TOH and 104 hips with AVN were retrospectively included. Femoral heads and necks with segmented radiomics features were extracted. Three ML classifiers (XGboost, CatBoost and SVM) using 38 relevant radiomics features were trained on 70% and validated on 30% of the dataset. ML performance was compared to two musculoskeletal radiologists, a general radiologist and two radiology residents. XGboost achieved the best performance with an area under the curve (AUC) of 93.7% (95% CI from 87.7 to 99.8%) among ML models. MSK radiologists achieved an AUC of 90.6% (95% CI from 86.7% to 94.5%) and 88.3% (95% CI from 84% to 92.7%), respectively, similar to residents. The general radiologist achieved an AUC of 84.5% (95% CI from 80% to 89%), significantly lower than of XGboost (p = 0.017). In conclusion, radiomics-based ML achieved a performance similar to MSK radiologists and significantly higher compared to general radiologists in differentiating between TOH and AVN.

Deep Learning for the Differential Diagnosis between Transient Osteoporosis and Avascular Necrosis of the Hip

“…The multi-institutional patient cohort utilized in this work was previously used for the development of radiomics signatures of TOH and AVN [ 7 ]. Study collection for dataset development was performed in an MSK radiology department specializing in bone marrow imaging and receiving domestic and international referrals for second opinions on complicated AVN and TOH cases.…”

“…All diagnostic decisions were discussed and reached in consensus with the referring surgeon. The MRI features used for the diagnosis of AVN and TOH have been described in detail previously [ 7 ]. Deep learning models were trained and validated with the use of mid-coronal STIR images through the femoral head and neck, since the pattern of BME on STIR images is sufficient for the imaging diagnosis of the two conditions in everyday MSK radiology practice; STIR is the sequence of choice for the depiction of bone marrow edema, which is a key feature of both conditions.…”

“…No pathophysiological similarity has been found between the two entities [ 5 ], and it has been demonstrated that patients with TOH do not progress to AVN, as previously thought [ 2 , 6 ]. Our group previously developed a radiomics–machine learning pipeline that allows differential diagnosis with an area under the curve (AUC) that is equal to MSK radiologists’ and better than general radiologists’ [ 7 ]. Accurate differentiation between the two entities is of the utmost importance, since it determines whether the patient will undergo conservative treatment for the management of TOH or surgical treatment (core decompression, vascularized fibular grafting, etc.)…”

“…Radiomics analysis such as the one presented in our previous work [ 7 ] has been widely used for the analysis of features derived from medical imaging [ 9 ]. Traditional radiomics include the segmentation of a region of interest, the extraction of radiomics features using established algorithms, and the use of machine learning techniques for the development of classifiers that allow accurate diagnosis of a disease [ 10 , 11 ].…”

Research in Musculoskeletal Radiology: Setting Goals and Strategic Directions

“…28 Differentiation between conditions related to bone marrow edema that may or may not lead to joint arthroplasty has also been achieved with radiomics and machine learning. 29 For bone and soft tissue sarcomas, radiomics has achieved accurate grading, 30 31 prediction of pretreatment response to chemotherapy, 32 and prediction of recurrence, metastasis, and overall survival, 30 thus facilitating efficient treatment planning. Finally, radiomics has allowed the differentiation between metastases and multiple myeloma in lytic spinal lesions.…”