HTN: Hybrid Transformer Network for Curvature of Cervical Spine Estimation

Many young people have suffered from cervical spondylosis in recent years due to long-term desk work or unhealthy lifestyles. Early diagnosis is crucial for curing spondylosis. The Cobb angle method the most common assessing spinal curvature. However, manually measuring time-consuming and heavily dependent on personal experience. In this paper, we propose a fully automatic system curvature X-rays using method, which can reduce workload of clinicians provide reliable basis surgery. Hybrid transformer network (HTN) blends self-attention mechanism, self-supervision learning, feature fusion. addition, new dataset proposed evaluate our method. Our model achieve SMAPE 11.06% significant Pearson correlation coefficient 0.9619 (p < 0.001) dataset. absolute difference between ground truth prediction obtained less than 2°, implying clinical value. Statistical analysis proves reliability estimation. To further prove validity HTN was also trained evaluated public AASCE MICCAI 2019 challenge experimental results show that comparable performance state-of-the-art methods, means measure neck entire spine.