Autoencoders for compression: applications in medical image analysis

Simple item page
dc.contributor.authorShaik, Ifthakaar
dc.date.accessioned2022-07-18T10:40:11Z
dc.date.available2022-07-18T10:40:11Z
dc.date.issued2021
dc.descriptionA research report submitted to the Faculty of Science, University of the Witwatersrand, Johannesburg, in partial fulfilment of the requirements for the degree of Master of Science by Coursework and Research Report, 2021en_ZA
dc.description.abstractWe propose the application of autoencoders to the task of compression within the field of blood analysis. The growing need for medical care and shortage of medical personnel requires a more decentralised approach to medical diagnostics. Remote di-agnostics in rural areas requires the ability to transmit digital medical image data of typically large file sizes over possible low bandwidth networks. The regularity found in medical samples may be exploited to allow for higher compression than generic com-pression algorithms while maintaining quality suitable for medical interpretation by human doctors. This paper explores the ability of a deep learning technique, namely autoencoders, to learn this regularity to create a lower-dimensional representation of the data. We assess the autoencoder’s performance by comparing it to traditional JPEG compression. We measure performance by considering the compression levels achieved and the quality of the resultant compressed images. Quality is measured using Peak Signal to Noise Ratio and Structural Similarity Index Measure, with the best techniques maximising values under each of these. Our results show that JPEG outperforms autoencoders at low compression levels, but as the compression level increases, autoencoders outperform JPEG based on the quality metrics used. We further test the impact of compression on the task of classification using a classification neural net. The results from our experiments show that compression negatively affects accuracy, however this loss in accuracy may be recovered by retraining the classifier on compressed images. The combination of results indicates that autoencoder-based compression techniques hold promising applications in the field of remote medical diagnostics, especially in rural areas where the ability to transmit data is limited by infrastructureen_ZA
dc.description.librarianCK2022en_ZA
dc.facultyFaculty of Scienceen_ZA
dc.identifier.urihttps://hdl.handle.net/10539/33026
dc.language.isoenen_ZA
dc.schoolSchool of Computer Science and Applied Mathematicsen_ZA
dc.titleAutoencoders for compression: applications in medical image analysisen_ZA
dc.typeThesisen_ZA
Files
Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
MSc_Research_Report___Ifthakaar_2021_Final.pdf
Size:
2.07 MB
Format:
Adobe Portable Document Format
Description:
Download
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.71 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
ETD Collection