Content-based Image Extraction and Analysis for Medical Radiographic Images

The amount of multimedia and visual data generated throughout the different fields is growing rapidly; a fact that has raised the need for a method which allows searching and accessing the provided data both quickly and easily, since querying data repositories must not be based solely on simple text. This paper presents the Medical Content-Based Image Retrieval (MCBIR) that introduces a unique image processing and computer vision system to the medical industry. The MCBIR Support System introduced herein is intended to be easy and user friendly; it also avails a decision support system and a built in knowledge-based system. These sub-systems should be able to help in providing a pre-diagnosis that is to support the decision made by the medical experts. As per the system’s implementation, there are three main steps: the extraction of random sub windows from medical images; the building of the ensembles of trees from the extracted sub windows; and, the derivation of similarity measures between images and their practical use in MCBIR. The proposed solution operates on the Indexing Random Sub Windows with randomized trees and a critical deduction on scaling found in A Generic Approach for Image Classification Based on Decision Tree Ensembles and Local Sub-Windows. However, advanced MCBIR faces some limitations such as the high cost required for production, and the need of training data to be used for machine learning. A solution for training data issue can be achieved by acquiring the available data of a certain medical institute. As for future developments, future modules can be introduced to the system by linking it to the Picture Archiving and Communication System (PACS) of the medical institute or hospital.