Automated Analysis of Digital Radiographs to Support an Image Audit
Abstract
In this paper, we detail how we used MATLAB (registered trademark) to analyze digital radiographs. Our dataset consisted of 153,516 images of M1 shells. We created a MATLAB (registered trademark) program that converted each image into a matrix of gray values corresponding to each pixel. We identified regions of interest by pixel location and performed our analysis using statistics associated with each region. The program successfully distinguished three different categories of defects, sorted the images according to which case they manifested, and had a category for ambiguous cases that required human attention. This methodology flags the most obvious cases, leaving the more contentious work for human eyes. It also allows for potential critical escapes to be quickly identified and brought to the attention of Level 3 radiographers and program managers. This paper will introduce the theory behind our method, explain the method used, and detail the development of the program. We will describe the performance of the program referring to the M1 dataset. We will then assess our results and speculate about future developments and applications.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 13, 2019
- Accession Number
- AD1080370
Entities
People
- Michael Cairo
- Walter Rose