Microcalcification detection in mammography for early breast cancer diagnosis using deep learning technique / Leong Yew Sum
Breast Cancer is one of the common cancers in women and may cause lives to be lost if they were misdiagnosed and left untreated. Existence of breast microcalcifications are common in breast cancer patients and they are an effective indicator of early breast cancer. This project will incorporate t...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://studentsrepo.um.edu.my/13611/ http://studentsrepo.um.edu.my/13611/1/Leong_Yew_Sum.jpg http://studentsrepo.um.edu.my/13611/8/yew_sum.pdf |
| Summary: | Breast Cancer is one of the common cancers in women and may cause lives to be lost if
they were misdiagnosed and left untreated. Existence of breast microcalcifications are
common in breast cancer patients and they are an effective indicator of early breast
cancer. This project will incorporate the use of machine learning in segmenting breast
mammogram images with calcifications of either benign or malignant cases for early
breast cancer diagnosis. ROI images of breast microcalcification will be utilized to train
several pretrained models from fastai library in Google Colaboratory platform using
supervised learning with a ratio of 0.80 for training dataset and 0.20 for validation
dataset. Image processing of ROI images were conducted to remove possible artifacts
and noises in order to enhance the quality of the images before training. The pretrained
models that were included in this study are Resnet34, Resnet50, VGG16 and Alexnet.
Different hyperparameters such as epoch, batch size etc were tuned in order to obtain
the best possible result in this study. Confusion matrices were utilized in order to
measure the output parameters of the models for comparison in terms of performance.
The result from this study shows that Resnet50 achieves the highest accuracy with a
value of 97.58%, followed by Resnet34 of 97.35%, VGG16 of 96.97% and finally
Alexnet of 83.06%. |
|---|