Otitis Media Analysis - An Automated Feature Extraction and Image Classification System
Kasher, Muhammad Shazam (2018)
Kasher, Muhammad Shazam
Metropolia Ammattikorkeakoulu
2018
All rights reserved
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-201805036257
https://urn.fi/URN:NBN:fi:amk-201805036257
Tiivistelmä
The main goal of the project was to develop novel Artificial Intelligence solutions using images and open source tools based on computer vision and deep learning to analyse and detect otitis media infection. The objective was to develop these solutions to help and support medical professionals in detecting otitis media at an early stage and making the final diagnosis.
Advanced computer vision, deep learning and transfer learning methods were used to develop two software systems. The first software system supports the professional by applying feature engineering to the images and extracting certain features that can help them in classifying the images manually. The second software system is an automated image classification software based on deep learning, which performs classification task for the professional which then can be used to make the final diagnosis.
The computer vision based system performed well on the available data and was successful in extracting all five key features which can be used to manually classify images. The deep learning based image classification software performed fairly good as well and achieved an accuracy of 82.2% with InceptionV3 architecture and 80% accuracy with MobileNets architecture.
It was concluded that the deep learning software system works better overall for classification tasks with high accuracy. In future development, a significantly higher accuracy can be reached using more data and more sophisticated algorithms.
Advanced computer vision, deep learning and transfer learning methods were used to develop two software systems. The first software system supports the professional by applying feature engineering to the images and extracting certain features that can help them in classifying the images manually. The second software system is an automated image classification software based on deep learning, which performs classification task for the professional which then can be used to make the final diagnosis.
The computer vision based system performed well on the available data and was successful in extracting all five key features which can be used to manually classify images. The deep learning based image classification software performed fairly good as well and achieved an accuracy of 82.2% with InceptionV3 architecture and 80% accuracy with MobileNets architecture.
It was concluded that the deep learning software system works better overall for classification tasks with high accuracy. In future development, a significantly higher accuracy can be reached using more data and more sophisticated algorithms.