«MedIm: Artificial Intelligence System for Pneumonia Detection»

AUTHOR: NARGIZ AKHMETOVA

Abstract

In contemporary society, despite the profound improvement in the radiology sphere of medicine, the difficulty is still presented by incompetence in interpreting the results of the diagnostics. The analysis of medical images is complicated by the complexity of the data itself, including 3D tomographic images based on different physical principles and predetermined by their characteristics and limitations. This fact bolsters the high risk of human factor error. The goal of the MedIm project is to bring automation to the detection of pneumonia based on X-ray images, which includes all the steps of diagnosis and final evaluation. Machine learning offers the ability to learn suitable models from previously analyzed data. Structured AI use cases that define specific parameters of datasets for training and testing algorithms, contribute to the development of multiple sites in developing training and validation datasets. This feature can help ensure that algorithms respect the technical, geographic, and demographic diversity in patient populations and acquisition of images, which subsequently results in higher accuracy of diagnosis evaluation.

To test the hypothesis that the implementation of the principles of artificial intelligence in pneumonia disease diagnosis will have a high level of efficiency, four machine learning classifier’s predicting accuracy were tested. After proper initialization, the neural networks were given identical data for training and testing, after which the calculation of the model's accuracy score was made. Based on the results of the calculation, it has been seen that the most efficient and functional library for X-ray image analysis is CNN Classifier. The accuracy of its forecast was 92%. Subsequently, the convolutional neural network was chosen as the main tool for analytics and forecasting the disease in the project’s final version.

The results of the research work suggest that the “MedIm” machine learning algorithm achieved its goal of identifying lung diseases using X-ray images, thereby confirming the hypothesis of the high efficiency of AI in diagnosing and interpreting radiological data.