Mammogram breast mass classification using deep convolutional neural network / Mohd Nafie Maslan
The aim of this study is to investigate a deep convolutional neural network (DCNN) image classification technique to classify breast mass (i.e., benign or malignant) on mammogram image. DCNN architecture model of AlexNet, GoogLeNet and VGG-16 was compared to evaluate the performance of classifying b...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://studentsrepo.um.edu.my/13614/ http://studentsrepo.um.edu.my/13614/1/Mohd_Nafie_Maslan.jpg http://studentsrepo.um.edu.my/13614/8/nafie.pdf |
| _version_ | 1848774865340858368 |
|---|---|
| author | Mohd Nafie, Maslan |
| author_facet | Mohd Nafie, Maslan |
| author_sort | Mohd Nafie, Maslan |
| building | UM Research Repository |
| collection | Online Access |
| description | The aim of this study is to investigate a deep convolutional neural network (DCNN) image classification technique to classify breast mass (i.e., benign or malignant) on mammogram image. DCNN architecture model of AlexNet, GoogLeNet and VGG-16 was compared to evaluate the performance of classifying breast mass patch. The dataset of breast mass was obtained from public database, Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM). Transfer learning which pre-trained the DCNN model on large-scale ImageNet, and data augmentation to increased number of data technique, were applied in this study. In total, 13,260 images of training dataset were imported into the NVIDIA deep learning GPU training system (DIGITS). A trained model of AlexNet, GoogLeNet and VGG-16 was created and selected based on the high accuracy and low loss. 540 images of testing dataset were applied to the trained DCNN model. The classification performance between the prediction and actual images were evaluated using confusion matrix and receiver operating characteristic (ROC) curve, to calculate accuracy, sensitivity/ recall, precision, specificity, F1-score and area under the curve (AUC). As a result; AlexNet and GoogLeNet have the highest specificity (71.48%); GoogLeNet have the highest accuracy (73.89%) and precision (72.79%); VGG-16 have the highest sensitivity/ recall (80%) and F1-score (75.26%). VGG-16 (0.8207) has the highest AUC, trailed by GoogLeNet (0.8064) abd AlexNet (0.7601). As conclusion, the overall ability of VGG-16 and GoogLeNet was superior over AlexNet in discriminated benign and malignant of breast mass patch. |
| first_indexed | 2025-11-14T14:05:05Z |
| format | Thesis |
| id | um-13614 |
| institution | University Malaya |
| institution_category | Local University |
| last_indexed | 2025-11-14T14:05:05Z |
| publishDate | 2022 |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | um-136142022-08-16T22:48:48Z Mammogram breast mass classification using deep convolutional neural network / Mohd Nafie Maslan Mohd Nafie, Maslan R Medicine (General) T Technology (General) The aim of this study is to investigate a deep convolutional neural network (DCNN) image classification technique to classify breast mass (i.e., benign or malignant) on mammogram image. DCNN architecture model of AlexNet, GoogLeNet and VGG-16 was compared to evaluate the performance of classifying breast mass patch. The dataset of breast mass was obtained from public database, Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM). Transfer learning which pre-trained the DCNN model on large-scale ImageNet, and data augmentation to increased number of data technique, were applied in this study. In total, 13,260 images of training dataset were imported into the NVIDIA deep learning GPU training system (DIGITS). A trained model of AlexNet, GoogLeNet and VGG-16 was created and selected based on the high accuracy and low loss. 540 images of testing dataset were applied to the trained DCNN model. The classification performance between the prediction and actual images were evaluated using confusion matrix and receiver operating characteristic (ROC) curve, to calculate accuracy, sensitivity/ recall, precision, specificity, F1-score and area under the curve (AUC). As a result; AlexNet and GoogLeNet have the highest specificity (71.48%); GoogLeNet have the highest accuracy (73.89%) and precision (72.79%); VGG-16 have the highest sensitivity/ recall (80%) and F1-score (75.26%). VGG-16 (0.8207) has the highest AUC, trailed by GoogLeNet (0.8064) abd AlexNet (0.7601). As conclusion, the overall ability of VGG-16 and GoogLeNet was superior over AlexNet in discriminated benign and malignant of breast mass patch. 2022-01 Thesis NonPeerReviewed application/pdf http://studentsrepo.um.edu.my/13614/1/Mohd_Nafie_Maslan.jpg application/pdf http://studentsrepo.um.edu.my/13614/8/nafie.pdf Mohd Nafie, Maslan (2022) Mammogram breast mass classification using deep convolutional neural network / Mohd Nafie Maslan. Masters thesis, Universiti Malaya. http://studentsrepo.um.edu.my/13614/ |
| spellingShingle | R Medicine (General) T Technology (General) Mohd Nafie, Maslan Mammogram breast mass classification using deep convolutional neural network / Mohd Nafie Maslan |
| title | Mammogram breast mass classification using deep convolutional neural network / Mohd Nafie Maslan |
| title_full | Mammogram breast mass classification using deep convolutional neural network / Mohd Nafie Maslan |
| title_fullStr | Mammogram breast mass classification using deep convolutional neural network / Mohd Nafie Maslan |
| title_full_unstemmed | Mammogram breast mass classification using deep convolutional neural network / Mohd Nafie Maslan |
| title_short | Mammogram breast mass classification using deep convolutional neural network / Mohd Nafie Maslan |
| title_sort | mammogram breast mass classification using deep convolutional neural network / mohd nafie maslan |
| topic | R Medicine (General) T Technology (General) |
| url | http://studentsrepo.um.edu.my/13614/ http://studentsrepo.um.edu.my/13614/1/Mohd_Nafie_Maslan.jpg http://studentsrepo.um.edu.my/13614/8/nafie.pdf |