Clinical evaluation of a commercially viable machine learning algorithm to automatically detect shoulder muscle pathology

项目总结 肩袖修复是进行次数最多的整形外科手术之一(美国每年有40万例手术 年)，但仍然是一个非常具有挑战性的临床问题。虽然肩袖的手术修复寻求改善 肩关节功能和稳定性，手术结果差异很大，因为，术前，很难 在目前的评估方法下，预测哪些患者将从手术中受益，哪些患者不会。 该项目的重点是开发独特的技术来取代现有的方法来生产快速、 准确评估能够大规模商业部署的旋转器袖带。 有重要的科学证据表明，肩袖肌肉的过度脂肪渗透和萎缩 导致不良结局，因为脂肪组织的存在限制了肌肉的恢复和 肌腱重建后再生。虽然目前的临床实践利用磁共振成像 (MRI)使用定性评分系统评估肩袖中的脂肪渗透，定性评分几乎没有- TO-O与脂肪渗透和萎缩的定量指标无相关性。纳入量化 测量将极大地改善临床治疗决策；然而，现有方法将 需要大量的人工输入，因此在临床上是不可行的。一种快速准确的车牌分割方法 肩袖肌肉和脂肪渗透对于改善结果和减少不必要的手术是必不可少的。 在该项目的第一阶段，我们成功地开发并验证了基于深度学习的 来自临床扫描的肩袖肌肉和脂肪渗透的自动算法。通过 创建一个广泛的数字数据库，包括健康和病理肩袖临床扫描，我们开发 一种新的方法来解释扫描覆盖的可变性，这导致了关键的旋转肌袖带的建立 可以快速、准确地从磁共振图像中得出肌肉指标。我们现在有了一个原型产品 这已经准备好进行Beta测试了。在第二阶段，我们建议进行一项前瞻性临床研究，以 确定哪些MRI得出的肌肉指标最能预测肩袖修复手术的结果。在……里面 目标1，我们将与多个骨科中心合作，对符合以下条件的肩袖进行术前分析 考虑进行肩袖修复手术，然后将手术前和手术后的指标联系起来 结果。在目标2中，我们将开发和改进用户界面以及最终将 部署用于临床使用。该项目的完成将使510(K)申请成为市场清算。这 该项目将显著提高肩部病理评估的准确性，从而促进诊断 和肩部疾病的治疗，改善昂贵的整形外科手术的结果，并有可能 甚至取消不必要的程序，所有这些都将改善患者护理并降低相关成本。

PROJECT SUMMARY Rotator cuff repairs are amongst the most performed orthopedic surgeries (>400,000 surgeries in the US per year) but remain a very challenging clinical problem. While surgical repair of the rotator cuff seeks to improve shoulder function and stability, the surgical outcomes vary significantly because, pre-operatively, it is difficult under current evaluative methods to predict which patients will benefit from surgery versus those who will not. The focus of this project is to develop unique technology that replaces current methods to produce a rapid, accurate assessment of rotator cuffs capable of large-scale commercial deployment. There is significant scientific evidence that excessive fat infiltration and atrophy of the rotator cuff muscles lead to poor outcomes because the presence of fatty tissue limits the ability for the muscle to recover and regenerate following tendon reconstruction. While current clinical practice utilizes magnetic resonance imaging (MRI) to evaluate fat infiltration in the rotator cuff using qualitative scoring systems, qualitative scoring has little- to-no correlation with quantitative measures of fat infiltration and atrophy. Incorporating quantitative measurements would dramatically improve clinical treatment decision-making; however, existing methods would require substantial manual input and thus is not clinically viable. A fast and accurate method for segmenting the rotator cuff muscles and fat infiltration is essential for improving outcomes and reducing unnecessary surgeries. During the Phase I period of this project, we successfully developed and validated a deep-learning-based automatic algorithm for quantification of rotator cuff muscle and fatty infiltration from clinical scans. Through the creation of an extensive digital database of both healthy and pathological rotator cuff clinical scans, we developed a novel method to account for variability in scan coverage, which led to the establishment of key rotator cuff muscle metrics that can be derived quickly and precisely from the MR images. We now have a prototype product that is ready for beta-testing. In the Phase II period, we propose to perform a prospective clinical study to determine which MRI-derived muscle metrics that best predict the outcomes of rotator cuff repair surgeries. In Aim 1, we will partner with multiple orthopedic centers to perform pre-operative analysis of rotator cuffs that are being considered for rotator cuff repair surgery, and then relate the pre-operative metrics with post-operative outcomes. In Aim 2, we will develop and refine the user interface and associated metrics that will be ultimately deployed for clinical use. Completion of this project will enable a 510(k) application for market clearance. This project will significantly improve the accuracy of shoulder pathology assessments, thus advancing the diagnosis and treatment of shoulder pathologies, improving the outcomes of costly orthopedic procedures, and potentially even eliminating unnecessary procedures, all of which will improve patient care and lower the associated costs.