Evaluation of inter- and intra-operator reliability of manual segmentation of femoral metastatic lesions

• Published in: International journal for computer assisted radiology and surgery 2021-10, Vol.16 (10), p.1841-1849
• Main Authors: Ataei, Ali, Eggermont, Florieke, Baars, Milan, van der Linden, Yvette, de Rooy, Jacky, Verdonschot, Nico, Tanck, Esther
• Format: Article
• Language: English
• Subjects: Biomechanics; Biomedical materials; Bone Neoplasms - diagnostic imaging; Bones; Computed tomography; Computer Imaging; Computer Science; Femur - diagnostic imaging; Health Informatics; Humans; Imaging; Lesions; Medicine; Medicine & Public Health; Metastasis; Original; Original Article; Osteoblasts; Pattern Recognition and Graphics; Radiology; Reliability analysis; Reproducibility of Results; Segmentation; Surgery; Tomography, X-Ray Computed; Vision
• ISSN: 1861-6410; 1861-6429
• DOI: 10.1007/s11548-021-02450-w

Purpose Accurate identification of metastatic lesions is important for improvement in biomechanical models that calculate the fracture risk of metastatic bones. The aim of this study was therefore to assess the inter- and intra-operator reliability of manual segmentation of femoral metastatic lesions. Methods CT scans of 54 metastatic femurs (19 osteolytic, 17 osteoblastic, and 18 mixed) were segmented two times by two operators. Dice coefficients (DCs) were calculated adopting the quantification that a DC˃0.7 indicates good reliability. Results Generally, rather poor inter- and intra-operator reliability of lesion segmentation were found. Inter-operator DCs were 0.54 (± 0.28) and 0.50 (± 0.32) for the first and second segmentations, respectively, whereas intra-operator DCs were 0.56 (± 0.28) for operator I and 0.71 (± 0.23) for operator II. Larger lesions scored significantly higher DCs in comparison with smaller lesions. Of the femurs with larger mean segmentation volumes, 83% and 93% were segmented with good inter- and intra-operator DCs (> 0.7), respectively. There was no difference between the mean DCs of osteolytic, osteoblastic, and mixed lesions. Conclusion Manual segmentation of femoral bone metastases is very challenging and resulted in unsatisfactory mean reliability values. There is a need for development of a segmentation protocol to reduce the inter- and intra-operator segmentation variation as the first step and use of computer-assisted segmentation tools as a second step as this study shows that manual segmentation of femoral metastatic lesions is highly challenging.