Lung Data Analysis With Deep Learning

Harikrishnan KESAVAN VIJAYAKUMAR

Bachelor's thesis

Abstract:

Medical images can have extremely high resolutions which cannot be handled properly by typical state­of­art machine learning models. In this thesis, I compared the performance of two approaches of multiple instance learning models where the high resolution images are downscaled into smaller patches and low dimensional embedding are calculated using Resnet. Then low dimensional embedding are aggregated using multiple instance learning to attain class labels. The data set for this thesis consisted of high resolution histological slides of human lung which were classified to contain cancer or not.

Abstract:

Medical images can have extremely high resolutions which cannot be handled properly by typical state­of­art machine learning models. In this thesis, I compared the performance of two approaches of multiple instance learning models where the high resolution images are downscaled into smaller patches and low dimensional embedding are calculated using Resnet. Then low dimensional embedding are aggregated using multiple instance learning to attain class labels. The data set for this thesis consisted of high resolution histological slides of human lung which were classified to contain cancer or not.

Keywords

CLAM, Multiple instance learning, Lung data analysis

Language used: English

Date on which the thesis was submitted / produced: 18. 11. 2021

Thesis defence

• Supervisor: prof. Dr. Sepp Hochreiter, MSc. Andreas Mitterecker

UNIVERSITY OF SOUTH BOHEMIA IN ČESKÉ BUDĚJOVICE

Faculty of Science

Bachelor programme / field:
Applied Informatics / Bioinformatics