Dipartimento di Informatica, Università degli Studi di Bari 'Aldo Moro', Bari, Italy.
Ospedale Policlinico San Martino, Genoa, Italy.
Int J Med Inform. 2019 Feb;122:13-19. doi: 10.1016/j.ijmedinf.2018.11.010. Epub 2018 Nov 30.
In recent years, cytological observations in the Rhinology field are being increasingly utilized. This development has taken place over the last two decades and has proven to be fundamental in defining new nosological entities and in driving changes in the previous classification of rhinitis. The simplicity of the technique and its low invasiveness make nasal cytology a practical diagnostic tool for all rhino-allergology services. Furthermore, since it allows the monitoring of responses to treatment, this method plays an important role in guiding a more effective and less expensive diagnostic program. Microscopic observation requires prolonged effort by a specialist, but the modern scanning systems for cytological preparations and the new affordable digital microscopes allow to design a software support system, based on deep learning techniques, to relieve specialist's tiring activity.
By means of the system presented in this paper, it is possible to automatically identify and classify cells present on a nasal cytological preparation based on a digital image of the preparation itself. Thus, an interesting diagnostic support has been made available to the rhino-cytologist, who can quickly verify that the cells have been correctly classified by the software system: any few unclassified or incorrectly classified cells can be quickly sorted by the specialist itself, then one or more diagnosis can be suggested by this system, taking into consideration also the anamnesis of each patient. The final diagnosis can be defined by the specialist, also based on the result of the prick test and the observation of the nasal cavity.
In the system presented herein, image processing and image segmentation techniques have been used to find images of cellular elements within the preparation. Cell classification is based on a convolutional neural network composed of three blocks of main layers. Cell identification (first step, image segmentation) exhibits sensitivity greater than 97%, while cell classification (second step, seven cytotypes) attained a mean accuracy of approximately 99% on the test set and 94% on the validation set.
This complete system supports clinicians in the preparation of a rhino-cytogram report.
近年来，鼻科学领域的细胞学观察越来越受到重视。这种发展是在过去二十年中发生的，它对于定义新的疾病实体和推动以前的鼻炎分类的改变至关重要。该技术的简单性及其微创性使其成为所有鼻过敏服务的实用诊断工具。此外，由于它可以监测治疗反应，因此该方法在指导更有效和更经济的诊断方案方面发挥着重要作用。显微镜观察需要专家进行长时间的努力，但是细胞学标本的现代扫描系统和新的经济实惠的数字显微镜允许设计基于深度学习技术的软件支持系统，以减轻专家的疲劳活动。
通过本文提出的系统，可以根据细胞学标本的数字图像自动识别和分类标本上存在的细胞。因此，为鼻细胞学专家提供了一种有趣的诊断支持，他们可以快速验证软件系统是否正确分类了细胞：软件系统可以快速由专家分类任何少量未分类或分类错误的细胞，然后系统可以根据每位患者的病史提出一种或多种诊断。最终诊断可以由专家根据皮试结果和鼻腔观察结果来定义。
在本文提出的系统中，使用图像处理和图像分割技术在标本中找到细胞元素的图像。细胞分类基于由三个主要层块组成的卷积神经网络。细胞识别（第一步，图像分割）的灵敏度大于 97%，而细胞分类（第二步，七种细胞类型）在测试集上的平均准确率约为 99%，在验证集上的准确率约为 94%。
这个完整的系统支持临床医生准备鼻细胞学报告。
