Machine learning-based quality control of canine thoracic radiographs

基于机器学习的犬胸部X光片质量控制

• 批准号: 560314-2020
• 负责人: Komeili, Amin
• 金额: $ 4.01万
• 依托单位: University of Guelph
• 依托单位国家: 加拿大
• 项目类别: Alliance Grants
• 财政年份: 2021
• 资助国家: 加拿大
• 起止时间: 2021-01-01 至 2022-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=723353
• 关键词: Machine; learning; based; quality; control

In animal hospitals and veterinary clinics, radiographs are taken by veterinary technicians and are often sent for a teleradiology consult by radiologists who are not present on-site. Turn around times for these studies range from 1 hour (STAT cases) to 2-3 days (non-emergency cases). Low-quality or poorly positioned radiographs may cause erroneous diagnosis and, in severe cases, may be classified as nondiagnostic by the radiologist. Therefore, quality control of the radiographs plays an important role in providing a reliable interpretation and convenient service to the patient. The training of personnel in proper positioning as well as modern x-ray units with wide exposure latitude and dynamic range help to reduce technical errors and increase quality. However, radiologists still receive radiographs with inappropriately collimated anatomy and inadequate positioning. As a result, there are a considerable number of canine thorax radiographs that render nondiagnostic or are limited in their diagnostic value. This increases the cost to veterinary clinics and clients, as well as increases radiation exposure to patients and personnel. Therefore, the importance of good image quality in radiographic studies to avoid the misleading or erroneous diagnosis, to reduce overall costs and to reduce radiation exposure, cannot be overstated. We propose a machine learning algorithm that analyzes the canine lateral thorax radiograph and evaluates its appropriateness before it is sent to the radiologist for diagnosis. We also aim to develop a sensor system that helps position the patient and reduce the room for technical error. The proposed work will benefit Canadian veterinarians as well as pet owners by 1) reducing the costs associated with repeating sedation and radiograph acquisition; 2) reducing the financial burden on the client to return to the veterinary clinic; 3) reducing technical staff availability; 3) eliminating roughly $2.5 variable cost per radiograph retake; and 4) eliminating the $50-100 telemedicine consultation fee per radiographic study if resubmitted for evaluation.

在动物医院和兽医诊所，X光照片由兽医技术人员拍摄，并经常由不在现场的放射科医生送去进行远程放射学咨询。这些研究的周转时间从1小时(统计病例)到2-3天(非紧急病例)不等。低质量或位置不佳的X光照片可能会导致错误诊断，在严重的情况下，可能会被放射科医生归类为非诊断。因此，对X线片的质量控制对于为患者提供可靠的解释和便捷的服务起着重要的作用。对适当位置的人员进行培训，以及配备具有宽曝光纬度和动态范围的现代X光机，有助于减少技术错误和提高质量。然而，放射科医生仍然会收到解剖结构不适当、定位不适当的X光片。因此，有相当数量的犬胸片没有诊断价值或诊断价值有限。这增加了兽医诊所和客户的成本，并增加了对患者和人员的辐射暴露。因此，在放射学研究中，良好的图像质量对于避免误诊或误诊、降低总体成本和减少辐射暴露的重要性怎么强调都不为过。我们提出了一种机器学习算法来分析犬的侧位胸片，并在将其发送给放射科医生进行诊断之前评估其适宜性。我们还计划开发一种传感器系统，帮助定位患者并减少技术错误的空间。拟议的工作将使加拿大兽医和宠物主人受益，因为1)减少了与重复镇静和获取X光照片相关的成本；2)减少了客户返回兽医诊所的财务负担；3)减少了技术人员的可获得性；3)每次重新拍摄X光照片可节省约2.5美元的可变成本；4)如果重新提交进行评估，则不再需要支付每项X光检查50-100美元的远程医疗咨询费。