Radiographic Dataset for VHS determination learning process

The data collected at baseline include latero lateral thoracic radiographic images of canine patients. This data was collected in 2019- 2020. The number of patients is 156 canine patients. The dataset consists of 156 radiographic images . The images are in PNG format.

Materials and Methods
Radiographic images were taken and collected at the Small Animal Veterinary Teaching Hospital, Veterinary Sciences Reasearch Institute, Autonomous University of Baja California. The image collection and classification was performed between 2019- 2020. The instrument used in the radiographic image adquisition process is Universal AV Choice X-Ray System and Rayence XMARU 1417PGA-PCA 14×17″ cassette digital detector (DR system). In the development of radiographic studies, techniques were used in the range of the following parameters using a high peak kilovoltage (kVp 70-80) and low milliampere-second technique (mA 200, ms 10-40) equivalent to 2-8 mAs. This technique allows for latitude (long gray scale) images, which are important when evaluating the structures of the thorax. For a right and left lateral image, patients were positioned on the table with the dependent side down. Relevant anatomical structures in the determination of VHS were included. The laterals beam projections extended from the cranial margin of the manubrium to the caudo-dorsal margin of the lung margin/diaphragmatic crus. The sternum of the patient was included so as not to exclude vital anatomy. The number of images collected initially was 628 and were stored in a DICOM format.
DICOM images were converted to PNG format by using a DICOM converter application included on the acquisition software. After performing classification and preprocessing, the number of images included in the final dataset was reduced to 152 representative images. All images were cropped to different sizes to remove unused and unimportant boundaries from the images using fast photo crop. As the original images contain confidential patient information not relevant for the overall educational purpose, the data was anonymized by removing all fields that would enable patient identification. The original VHS measurement was not included in the image because they may affect the output results of the training process.

基线阶段收集的数据包含犬类患者的胸侧位X光摄影影像（laterolateral thoracic radiographic images）。该数据集采集于2019至2020年，共纳入156只犬类患者，包含156张X光影像，影像格式为PNG。

材料与方法
X光影像采集于下加利福尼亚自治大学兽医科学研究所小型动物兽医教学医院，影像采集与分类工作开展于2019至2020年。本次影像采集所用设备为Universal AV Choice X射线系统（Universal AV Choice X-Ray System）与Rayence XMARU 1417PGA-PCA 14×17英寸暗盒式数字探测器（DR系统，DR system）。
在开展X光摄影检查时，采用的技术参数范围如下：高千伏管电压峰值（kVp 70-80）、低毫安秒曝光参数（管电流mA 200，曝光时长ms 10-40，对应总曝光量2-8 mAs）。该成像技术可获得宽灰阶（长灰阶）影像，这对胸部结构的评估具有重要意义。
针对左右侧位影像的拍摄，患者取检查台侧卧，受检侧朝下。影像纳入用于椎心评分（VHS）判定的相关解剖结构，侧位射线投射范围从胸骨柄颅侧缘延伸至肺缘/膈脚的背侧后缘，同时包含患者胸骨，以避免遗漏关键解剖结构。
初始采集的影像共计628张，存储格式为DICOM（DICOM）。随后通过采集软件内置的DICOM转换工具将影像转换为PNG格式。经分类与预处理后，最终数据集纳入的代表性影像缩减至152张。所有影像均通过快速照片裁剪工具（fast photo crop）进行不同尺寸的裁剪，以去除图像中无用且无关的边界区域。鉴于原始影像包含与整体教育研究目的无关的患者隐私信息，研究人员通过移除所有可识别患者身份的字段完成数据匿名化处理。此外，原始的VHS测量值未纳入影像，以避免其对模型训练的输出结果产生干扰。