A Survey on the Machine Learning Techniques for Automated Diagnosis from Ultrasound Images

Medical diagnostic systems has recently been very popular and reliable because of possible automatic detections. The machine learning algorithm is evolved as a core tool of computer-aided diagnosis (CAD) for automatic early and accurate disease detections. The algorithm follows region of interest (ROI) selection followed by specific feature extractions and selection from medical images. The selected features are then fed to suitable classifiers for disease identification. The machine learning algorithm's performance depends on the features selected and the classifiers employed for the job. This paper reviews different feature extraction selection and classification techniques for CAD from ultrasound images. Ultrasonography (USG), due to its portability and its non-invasive nature, is the prime choice of doctors for prescribing as an imaging test. A survey on the USG imaging based on four major diseases is performed in this paper, whose diagnosis followed by automatic detection. Various techniques applied for feature extraction, selection, and classification by different authors to achieve improved accuracy are tabulated. For medical images, we found texture based gray-level extracted features and SVM (support vector machine) classifiers to be more significant in improving classification accuracy, even achieving 100% accuracy in many research articles. However, many research articles also suggest the importance of student's t-test in improving classification accuracy by selecting significant features from extracted features. The proposed algorithm's accuracy also depends on the quality of medical images, which are frequently degraded by the introduction of noise and artifacts while imaging acquisition. So, challenges in denoising are added in this paper as a separate topic to highlight the role of the machine learning algorithm in removing noise and artifacts from the USG images.