Breast tumour segmentation using convolutional neural network on 3D computed tomography images
Image Segmentation of CT Images have always been costly in terms of time and money. The usual procedure includes having an experienced radiologist looking through the patient’s CT Images and manually segmenting out the cancer tumour. With the rise of computational power made available through advanc...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project (FYP) |
| Language: | English |
| Published: |
Nanyang Technological University
2021
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/149118 |
| _version_ | 1826112276711604224 |
|---|---|
| author | Ng, Yew Kong |
| author2 | Cai Yiyu |
| author_facet | Cai Yiyu Ng, Yew Kong |
| author_sort | Ng, Yew Kong |
| collection | NTU |
| description | Image Segmentation of CT Images have always been costly in terms of time and money. The usual procedure includes having an experienced radiologist looking through the patient’s CT Images and manually segmenting out the cancer tumour. With the rise of computational power made available through advancements in technology, more and more healthcare practitioners are looking into using Artificial Intelligence (AI) solutions to help solve such problems. We are able to observe more and more AI in healthcare research being conducted, especially during the recent 2-3 years.
As Deep Learning techniques in Computer Vision mature, it presents itself as a suitable candidate for use in medical diagnosis. Moreover, Deep Learning models have shown to even perform at radiologists’ level of performance at the segmentation task [35]. For this project, deep learning techniques will be applied to perform automatic tumour segmentation, which aims to reduce the manpower requirements as well as time needed for cancer diagnosis. We strive to accomplish this through the use of the U-Net architecture and its variants, which is widely cited in the medical image field for having relatively high accuracy & computational speed.
We are able to achieve a Dice score of 0.8790 on the test set using U-Net with MobileNetV2 encoder on custom loss with data augmentation. Almost all models trained on custom loss are able to achieve a Dice score of above 0.8 which show great promise of using AI to aid in faster diagnosis. |
| first_indexed | 2024-10-01T03:04:19Z |
| format | Final Year Project (FYP) |
| id | ntu-10356/149118 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T03:04:19Z |
| publishDate | 2021 |
| publisher | Nanyang Technological University |
| record_format | dspace |
| spelling | ntu-10356/1491182021-05-14T06:41:47Z Breast tumour segmentation using convolutional neural network on 3D computed tomography images Ng, Yew Kong Cai Yiyu School of Mechanical and Aerospace Engineering MYYCai@ntu.edu.sg Engineering Engineering::Mechanical engineering Image Segmentation of CT Images have always been costly in terms of time and money. The usual procedure includes having an experienced radiologist looking through the patient’s CT Images and manually segmenting out the cancer tumour. With the rise of computational power made available through advancements in technology, more and more healthcare practitioners are looking into using Artificial Intelligence (AI) solutions to help solve such problems. We are able to observe more and more AI in healthcare research being conducted, especially during the recent 2-3 years. As Deep Learning techniques in Computer Vision mature, it presents itself as a suitable candidate for use in medical diagnosis. Moreover, Deep Learning models have shown to even perform at radiologists’ level of performance at the segmentation task [35]. For this project, deep learning techniques will be applied to perform automatic tumour segmentation, which aims to reduce the manpower requirements as well as time needed for cancer diagnosis. We strive to accomplish this through the use of the U-Net architecture and its variants, which is widely cited in the medical image field for having relatively high accuracy & computational speed. We are able to achieve a Dice score of 0.8790 on the test set using U-Net with MobileNetV2 encoder on custom loss with data augmentation. Almost all models trained on custom loss are able to achieve a Dice score of above 0.8 which show great promise of using AI to aid in faster diagnosis. Bachelor of Engineering (Mechanical Engineering) 2021-05-14T06:41:46Z 2021-05-14T06:41:46Z 2021 Final Year Project (FYP) Ng, Y. K. (2021). Breast tumour segmentation using convolutional neural network on 3D computed tomography images. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/149118 https://hdl.handle.net/10356/149118 en A089 application/pdf Nanyang Technological University |
| spellingShingle | Engineering Engineering::Mechanical engineering Ng, Yew Kong Breast tumour segmentation using convolutional neural network on 3D computed tomography images |
| title | Breast tumour segmentation using convolutional neural network on 3D computed tomography images |
| title_full | Breast tumour segmentation using convolutional neural network on 3D computed tomography images |
| title_fullStr | Breast tumour segmentation using convolutional neural network on 3D computed tomography images |
| title_full_unstemmed | Breast tumour segmentation using convolutional neural network on 3D computed tomography images |
| title_short | Breast tumour segmentation using convolutional neural network on 3D computed tomography images |
| title_sort | breast tumour segmentation using convolutional neural network on 3d computed tomography images |
| topic | Engineering Engineering::Mechanical engineering |
| url | https://hdl.handle.net/10356/149118 |
| work_keys_str_mv | AT ngyewkong breasttumoursegmentationusingconvolutionalneuralnetworkon3dcomputedtomographyimages |