| |
|
| |
Aurthors/المؤلفون
Abstract/الملخص
Keywords/الكلمات المفتاحية
Content/أقسام الملف
Introduction
The importance of the research and its objectives
Research methods and materials
Literature Review
References |
| Breast cancer detection using deep learning |
| |
| |
| Dr. Ghada Saad(1) Aous Mohammad*(2) Haidar Haidar(3) |
1. phd in the Department of Biomedical Engineering of faculty of Mechanical and electrical Engineering - Tishreen University.
2. Master's in the Department of Computer and Automatic control Engineering of faculty of Mechanical and electrical Engineering - Tishreen University.
3. Master's in the Department of Computer and Automatic control Engineering of faculty of Mechanical and electrical Engineering - Tishreen University. |
| Email of the corresponding: aosmohamed93@gmail.com |
| |
| |
|
|
Abstract
Breast cancer is considered one of the most prevalent diseases in the world, and it is a disease that affects women greatly, its spread in recent years has led to an increase in death rates, and there is a possibility of an increase in the number of infections in the coming years, and given the problems that visual diagnosis of these diseases suffers from, including obtaining wrong and inappropriate results It is accurate, and it takes a lot of time and effort. Modern technologies, specifically deep learning techniques, have recently been used to analyze and classify these diseases in order to help doctors in early and accurate detection of diseases. In this research, a CNN model was built and the model was trained on a data set from Kaggle that contains On a variety of Ultrasound images of benign, malignant, and normal tumors of the breast, then a comparison was made between several pre-trained models, where the proposed model achieved the best assessment accuracy than the pre-trained models, and the results gave an accuracy rate of 97%, a sensitivity of 97%, and a specificity of 98%, and thus the results show the effectiveness The proposed algorithm as an aid model for clinicians. |
|
|
|
| Keywords: convolutional neural networks, benign, malignant, breast cancer, Ultrasound. |
|
| |
|
| |
|
| Introduction |
|
Cancer is the second leading cause of death worldwide in 2020, causing nearly 10 million deaths. Time among all types of cancer. Breast cancer is the most common among women worldwide, according to World Health Organization (WHO) figures, causing 685,000 women to die in 2020 [1]. Cancer arises when abnormal body cells begin to separate and come into contact with normal cells. And to make it malignant, the part of the body that contains breast cancer is the glands and milk ducts.
There are many diagnostic tests for breast cancer that are performed to confirm whether a person has a benign or malignant tumor, such as x-ray imaging, magnetic resonance imaging, tissue image analysis, mammography, and ultrasound [2], and visual diagnosis of these cancerous tumors sometimes leads to wrong or inaccurate results. As a result of human errors resulting from deficiencies and visual problems, the task of detecting and identifying cancer is a difficult task that takes a lot of time and is performed by experienced radiologists and specialized doctors, as the accuracy of the diagnosis depends on the doctor’s experience only, so the use of information technology has become necessary to overcome these limitations, which help doctors in detecting and diagnosing breast cancer.
Therefore, there was a need for CAD computer-aided detection systems that use artificial intelligence techniques to provide accurate diagnosis, as these systems help detect breast cancer at an early stage and provide better treatment and thus will increase the survival rate [3]. |
|
| |
|
| The importance of the research and its objectives |
|
Doctors need new and advanced methods to help them in early detection of diseases and reduce the possibility of disease spread and patient health deterioration. Using the convolutional neural networks model will lead to more accurate results in detecting breast cancer and helping doctors in early diagnosis and thus avoid diagnostic errors caused by deficiencies and human errors.
The research aims to develop an artificial intelligence model based on multi-classification convolutional neural networks for the diagnosis of breast cancer (benign-malignant-normal). |
|
| |
|
| Research methods and materials |
|
This research was carried out using the python programming language within the pycharm environment on a personal computer that has the specifications of an intel core i5-8250u CPU at 1.80 GHz and 8GB of system memory.
The research includes several stages starting from image acquisition up to the diagnosis stage by building a deep learning model that depends on CNN networks with the use of Keras and Tensorflow libraries to build the model. Figure (1) shows the block diagram of the work stages during this research. |
|
| |
|
 |
|
| Figure 1. A box diagram of the stages of work during this research. |
|
| Literature Review |
|
There are many previous studies that classified breast cancer, we will present in the following the most important of these studies in recent years.
Karthiga, R., and K. Narashimhan. [4] They presented a study that indicated that adjusting parameters and initializing weights is a major task for adapting pre-trained convolutional neural network models. The transfer learning method was used using the Alexnet and VGG16 models. The DCNN model was also developed and its performance was compared with pre-trained models on a data set. Kaggle, after training the model, the DCNN model achieved higher accuracy than the other models used in this study, achieving an accuracy rate of 93.38% in binary classification and an average accuracy of 89.29% in multiple classification.
Both Lévy and others [5] used a combination of learning transfer techniques, accurate pre-processing, and the use of virtual augmentation of data, in order to deal with the limitations of the data used represented in the data set Digital Database For Screening Mammography. The study used three models, namely Baseline, AlexNet, and GoogleNet. The GoogleNet model has an accuracy of 92%, followed by the AlexNet model, which has an accuracy of 89%, and finally the Baseline model, which has an accuracy of 60%.
In a study conducted by Yu, S., Liu et al. [6] using pre-trained deep neural networks (DCNNs) in the diagnosis of breast cancer radiographs from the Breast Cancer Digital Medical Repository (BCDR), the results of GoogleNet, AlexNet and shallow CNNs (CNN2) were compared. ,CNN3), experimental results indicate that GoogleNet achieved the best performance with an accuracy of 81%, followed by AlexNet with an accuracy of 79%, and finally CNN3, which gave an accuracy of 73%. |
|
| |
|
| References |
|
| |
|
1. World health organization (WHO). [Internet]. Cancer; [cited 2020]. Available from:https://www.who.int/ar/news-room/fact-sheets/detail/cancer.
2. Ghada S. Ahmad K. Qosai K.” ANN and Adaboost application for automatic detection of microcalcifications in breast cancer” , ScienceDirect; 2016: 5(9): pp:1-12.
3. David A. Omondiagbe et al (2019). Machine Learning Classification Techniques of Breast Cancer Diagnosis IOP Conf. Ser: Mater. Sci. Eng. 495 012033.
4. Karthiga, R., and K. Narashimhan. "Deep convolutional neural network for computer -. aided detection of breast cancer using histopathology images." Journal of Physics: conference Serie. Vol. 1767. No. 1. IOP Publishing, 2021.
5. Lévy, Daniel, and Arzav Jain. "Breast mass classification from mammograms using-18(2016). arXiv preprint arXiv:1612.00542 deep convolutional neural networks. |
|
| |
|
| |
|
| |
|
|
