Aiding Diagnosis and Classifying of Early Stage Osteonecrosis of the Femoral Head with Convolutional Neural Network Based on Magnetic Resonance Imaging

• DOI: 10.1007/s43465-024-01272-7
• Publication Date: 2024-12-04T07:16:27Z
• Authors: Chen Liang; Yingkai Ma; Xiang Li; Yong Qin; Minglei Li; Chuanxin Tong; Xiangning Xu; Jinping Yu; Ren Wang; Songcen Lv; Hao Luo
• Keywords: 03 medical and health sciences; 0302 clinical medicine; Original Article

Abstract Introduction The Steinberg classification system is commonly used by orthopedic surgeons to stage the severity of patients with osteonecrosis of the femoral head (ONFH), and it includes mild, moderate, and severe grading of each stage based on the area of the femoral head affected. However, clinicians mostly grade approximately by visual assessment or not at all. To accurately distinguish the mild, moderate, or severe grade of early stage ONFH, we propose a convolutional neural network (CNN) based on magnetic resonance imaging (MRI) of the hip joint of patients to accurately grade and aid diagnosis of ONFH.

Materials and Methods T1-MRI images of patients diagnosed with early stage ONFH were collected. Three orthopedic surgeons selected 261 slices containing images of the femoral head and labeled each case with the femoral head necrosis classification. Our CNN model learned, trained, and segmented the regions of femoral head necrosis in all the data.

Results The accuracy of the proposed CNN for femoral head segmentation is 97.73%, sensitivity is 91.17%, specificity is 99.40%, and positive predictive value is 96.98%. The diagnostic accuracy of the overall framework is 90.80%.

Conclusions Our proposed CNN model can effectively segment the region where the femoral head is in MRI and can identify the region of early stage femoral head necrosis for the purpose of aiding diagnosis.