支气管分级识别数据

用于结合支气管镜进行人体内支气管的分级识别，通过深度学习构建的模型快速识别支气管的分级分类，识别记录病灶发生的支气管位置，极大的提高了医生的工作效率，显著降低医疗事故。本数据和算法仅针对人体内的支气管视频或图像画面显示，不含个人数据、公共数据，无数据标识体现。本算法是对支气管中的不同分级的支气管进行识别和判断。对于图像识别进行训练，得到的估计误差值越小，识别的越准确。得到估计误差值的算法是在实例视频中，第一次出现置信度大于thresh（阈值）的图像时，记录当前矩形框的横向宽度L1。下一个时间节点（一般在上一次数据处理完时触发）记录矩形框的横向宽度L2，依次记录L3、L4、Lk一直记录到不识别或者置信度低。取当前的平均值Xk为估计的真实数据，根据卡尔曼增益的算法公式得出卡尔曼增益值Kk。在本系统中采用第一次测量的3%为测量误差E，估计误差e=（1-Kk）*E。此时得到当前支气管平均像素值，在像素值判断趋近于该值时，判断为预定义的支气管等级。

This deep learning-based model and supporting dataset are designed for hierarchical recognition of human bronchi during bronchoscopic examinations. It enables rapid hierarchical classification of bronchi, identifies and records the bronchial positions where lesions occur, thereby greatly improving clinicians' work efficiency and significantly reducing medical adverse events.
This dataset and corresponding algorithm are exclusively applicable to video or image frames depicting human bronchi, containing neither personal data nor public data, and do not include any data identification markers.
The algorithm is designed to recognize and classify bronchi of different hierarchical grades within the bronchial tree.
Training for the image recognition task follows the principle that a smaller estimated error value corresponds to more accurate recognition performance.
The algorithm for calculating the estimated error operates as follows: In the input video, when an image with a confidence score greater than the threshold thresh appears for the first time, record the horizontal width L1 of the current bounding box. At the subsequent time node (usually triggered upon completion of the previous data processing), record the horizontal width L2 of the bounding box, and sequentially record L3, L4, ..., Lk until recognition fails or the confidence score is too low. Take the average value Xk of these recorded widths as the estimated true data, and calculate the Kalman gain Kk using the formal Kalman gain algorithm formula. In this system, the measurement error E is defined as 3% of the first measured width, and the estimated error e = (1 - Kk) * E. At this point, the average pixel value of the current bronchus is obtained; when the pixel value of the input frame is close to this average value, the input is classified into a pre-defined bronchial grade.

搜集汇总

数据集介绍

特点

支气管分级识别数据是由浙江优亿医疗器械股份有限公司提供的企业数据，包含50条记录，用于结合支气管镜进行人体内支气管的分级识别，通过深度学习模型提高医生的工作效率和降低医疗事故。