Chestx-Edge: An Integrated Approach For Outlier Detection In Chest X-Rays Using Dimension Reduction And Edge Detection

The advancement of artificial intelligence technology has driven artificial intelligence research in many field s such as medicine. Medical intelligence focuses on co mputer- aided diagnosis , which uses signals from the human b ody to predict health problems and diagnose diseases t hrough medical examinations . However, processing a nd recognition of high- quality medical images requires expensive equipment and high energy consumption. To solve these problems, this paper proposes a new approach to detect outliers in chest radiographs using dimensionality reduction and edge detection techniques. The proposed method involves scanning X- ray images using a fixed-size window, performing differential imaging on adjacent segment images, and extracting edge information in binary format using AND operations. Subsequently, the extracted edge data is convolved using a detection filter to generate a series of lines that are divided into 16 types. The frequency of each line type is then summed to form a one- dimensional array for each segment image. This reduced data is used as input for a training model based on a recurrent neural network.To evaluate the effectiveness of the proposed model, se veral experiments were conducted using data showing COVID in chest X-ray. The results showed that the special education con solidation application achieved the highest accuracy a t 97.5%.