Javascript must be enabled for the correct page display

Breast Density Estimation in T1-weighted MRI using Deep Learning

Muskan, Muskan (2023) Breast Density Estimation in T1-weighted MRI using Deep Learning. Master's Thesis / Essay, Artificial Intelligence.


Download (13MB) | Preview
[img] Text
Restricted to Registered users only

Download (163kB)


Breast density is a significant risk factor for developing breast cancer, the most commonly diagnosed cancer in women. The screening for breast cancer is done predominantly using mammography, resulting in 3D mammograms. However, when the breast density is relatively high, the sensitivity of mammography decreases, which is when a breast MRI performs relatively well. Breast density estimation is challenging, and MRI as a 3D modality can provide more accurate density measurements. Researchers have proposed various methods for breast density estimation. However, a few of them have employed deep learning on breast MRIs. This study aims to perform breast density classification using deep convolutional neural networks (CNNs) on breast MRI. Additionally, it aims to learn the impact of different architectural choices such as using 3-dimensional vs 2-dimensional CNNs, giving higher weights to less represented classes (such as extremely dense), and the impact of adding transfer learning on the classification performance. A total of 960 breast MRI images were evaluated from 508 patients with a mean age of 45 years ± 11. The input to the deep learning pipeline was T1-weighted sequences from MRI images. Various ResNet-18 architectures were experimented with, such as 3D ResNet-18 with 3 MRI slices, or with the whole volume, 2D ResNet-18 with pre-trained ImageNet weights, and class weights. Three out of four models gave a reasonable agreement with the radiologists' assessments.

Item Type: Thesis (Master's Thesis / Essay)
Supervisor name: Dhali, M.A. and Ooijen, P.M.A. van and Jing, X.
Degree programme: Artificial Intelligence
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 16 May 2023 10:33
Last Modified: 16 May 2023 10:33

Actions (login required)

View Item View Item