Reviewing Machine Learning and Image Processing Based Decision-Making Systems for Breast Cancer Imaging
Hasnae Zerouaoui1, Ali Idri2,3
1Modeling, Simulation and Data Analysis, Mohammed VI Polytechnic University, Benguerir, Morocco.
Journal of Medical Systems
|January 6, 2021
Summary
This review analyzes machine learning and image processing for breast cancer (BC) detection in medical imaging. Deep learning excels in classification tasks, with mammography being the primary imaging modality.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Oncology
Background:
- Breast cancer (BC) is a leading cause of mortality in women globally, particularly those over 40.
- Medical image analysis offers crucial support for diagnosing and managing diseases like BC.
- Machine Learning (ML) and Image Processing (IP) are pivotal in advancing BC imaging analysis.
Purpose of the Study:
- To conduct a Structured Literature Review (SLR) on the application of ML and IP techniques in breast cancer imaging.
- To analyze trends and methodologies in BC imaging research from 2000 to August 2019.
- To identify prevalent medical tasks, ML techniques, datasets, and IP methods used in BC image analysis.
Main Methods:
- A systematic review of 530 papers published between 2000 and August 2019.
- Analysis based on ten criteria: publication details, study type, medical task, ML techniques, datasets, validation, performance, and IP methods (pre-processing, segmentation, feature extraction/selection).
- Categorization of ML techniques, focusing on Deep Learning (DL).
Main Results:
- Diagnosis emerged as the most frequent medical task, with Deep Learning (DL) predominantly used for classification.
- Classification was the primary ML objective, followed by prediction and clustering.
- Mammography was the most common imaging modality, with the MIAS dataset frequently utilized.
- Image pre-processing (noise reduction, normalization) and segmentation (thresholding) were widely applied.
- Both classical and DL methods were employed for feature extraction, with limited use of feature selection techniques.
Conclusions:
- ML and IP, especially DL, are integral to breast cancer imaging analysis, primarily for diagnosis and classification.
- Mammography remains the key imaging modality, and standardized datasets like MIAS are crucial for research.
- Further research could explore advanced feature selection and diverse imaging modalities for improved BC detection.
