China
Africa
Explore chapters and articles related to this topic
Performance Analysis of Deep Convolutional Neural Networks for Diagnosing COVID-19: Data to Deployment
Published in K. Gayathri Devi, Kishore Balasubramanian, Le Anh Ngoc, Machine Learning and Deep Learning Techniques for Medical Science, 2022
Recent progress in machine learning algorithms that have resulted in better performance can be attributed to the use of convolution layers, which can correctly identify despite variations in the image. Neural Networks are the types of machine learning algorithms that don’t need to be programmed with a clear set of distinct rules describing what to do with the input data. The neural network instead learns from processing many labeled examples that are provided to it during training to learn, what attributes of the input are necessary to provide the correct output [1]. Once a sufficient number of examples have been processed, the neural network can initiate to process new, unobserved inputs and successfully return precise results. Convolutional Neural Network (CNN) is a feature-based machine learning algorithm that is mainly used in classification problems [2]. Since CNN requires minimum pre-processing, it is widely preferred over other classification algorithms. By applying proper filters, spatial and temporal dependencies in the image can be captured which results in an efficient classification. The performance analysis of machine learning algorithms to diagnose COVID-19 is discussed in this chapter.
Comparison of 2D and 3D U-Nets for Organ Segmentation
Published in Jinzhong Yang, Gregory C. Sharp, Mark J. Gooding, Auto-Segmentation for Radiation Oncology, 2021
Convolutional neural networks (CNNs) have become one of the most successful tools in the field of medical image processing. In recent years, typical CNNs originating from the field of computer vision [1–5] have found their application in segmenting medical images for quantitative analysis [6–9]. Among these methods, fully convolutional networks (FCNs) [10,11] have been used for pixel-wise image segmentation. FCNs are only built using convolution, pooling, and activation layers, and use a patch-wise sliding window on the original input images. A detailed exploration of deep-learning architectures is given in Chapter 7. One of the drawbacks of FCNs is that segmentation must be performed for each pixel for a given position (center) of the sliding window. This was resolved by the U-net architecture [12], which modifies the FCN using an encoder and decoder structure that yields better segmentation in both digital natural image and medical image segmentation. The U-net has a symmetric multi-resolution structure and skips connections between the down-sampling path and the up-sampling path at each resolution. These skip connections combine local information at each resolution with the global information from the up-sampling blocks. The network has a large number of feature maps in the up-sampling path because of its symmetry. Therefore, it is sometimes necessary to reduce the number of channels in subsequent convolution layers to reduce memory usage.
Medical Images of Breast Tumors
Published in Abdel-Badeeh M. Salem, Innovative Smart Healthcare and Bio-Medical Systems, 2020
Yuriy Zaychenko, Galib Hamidov
The most part of last papers referring to the field of breast cancer classification were oriented on integer image (Doyle et al., 2008; Singh et al., 2015; Zhang et al., 2013; Zhang, Zhang, Coenen, Xiau, & Lu, 2014). Widespread implementation of breast image classification (BIC) and other forms of digital pathology, however, face barriers such as high cost of implementation, insufficient productivity compared to the amount of clinic procedures, interior technological problems, and opposition from pathologists and anatomists. Until now, most of the works based on histology breast cancer analysis were performed on small data sets. Some improvement in medical images data sets presented data set with 7,909 breast images obtained from 82 patients (Spanhol Oliveira, Petitjean, & Heutte, 2016). In this research, the authors estimated various texture descriptors and various classifiers and carried out their experiments with 82%–85% accuracy. The alternative to this approach is the application of convolutional neural network (CNN) for medical images processing and diagnostics, which is considered and developed in the present research.
The efficiency of artificial intelligence methods for finding radiographic features in different endodontic treatments - a systematic review
Published in Acta Odontologica Scandinavica, 2023
Shaqayeq Ramezanzade, Tudor Laurentiu, Azam Bakhshandah, Bulat Ibragimov, Thomas Kvist, EndoReCo , Lars Bjørndal
This systematic review revealed that most of the studies investigating the use of AI in endodontology were based on ANN or convolutional neural networks (CNN). Whenever there was a comparison between machine performance and human experts, the AI demonstrated good and often better accuracy than human performance alone, but whether it will benefit within a clinical setting is far too early to conclude. Ten of the included studies were on detecting periapical radiolucencies, three on detecting vertical root fracture, four on the classification of root morphology, four on pulp cavity segmentation, two on locating minor apical foramen and one on predicting outcomes involving radiographs in relation to endodontic retreatment. While the reported accuracy measurements seem promising, the quality of papers is considered low, as almost 60% of papers had some level of bias (Table 5). In particular, the reference standard domain was scored low in 37.5% of the papers.
Integrating artificial intelligence into an ophthalmologist’s workflow: obstacles and opportunities
Published in Expert Review of Ophthalmology, 2023
Priyal Taribagil, HD Jeffry Hogg, Konstantinos Balaskas, Pearse A Keane
There are many machine learning algorithms used in healthcare such as linear regression, logistic regression, decision trees, and random forests. None of these examples represent deep learning, which is a subset of machine learning that uses multiple layers of nodes connected in a neural network. These neural networks are able to process multiple data items, whilst preserving spatial distribution [7]. The incorporation of hidden layers aids the exploration of more complex non-linear data patterns [8]. Convolutional Neural Networks (CNN) are a subtype of neural networks used commonly in image recognition. By using multiple convolutional layers, they are able to process both simple and complex features (edges, lines, colors, shapes, etc.). Current examples of CNNs include AlexNet, GoogleNet, and ResNet[9]. Much of the successes in deep learning have been driven by CNNs.
Deep learning for assessing liver fibrosis based on acoustic nonlinearity maps: an in vivo study of rabbits
Published in Computer Assisted Surgery, 2022
Jinzhen Song, Hao Yin, Jianbo Huang, Zhenru Wu, Chenchen Wei, Tingting Qiu, Yan Luo
Machine learning is one of the prime branches of artificial intelligence, which has the ability to learn and improve from experience [20]. Deep learning technique, as one of machine learning technique, has attracted significant attention in machine learning field and obtained spectacular achievement, such as computer assisted diagnosis for tumor. Convolutional neural networks (CNN) is a widely investigated model of deep learning, which has been proven efficient in medical imaging recognition, especially in the fields of differential diagnosis between benign and malignant tumor and stroke [20–22]. Prediction of liver fibrosis stages is also a field in which CNN had been applied recent years. CNN using conventional B-mode ultrasound images and elastography images as input data displayed good diagnostic performance for liver fibrosis [23]. Hidden layers embedded in CNN automatically extract features that may not be detectable in humans’ naked eyes [24]. However, to our knowledge, there is no study applying CNN methods on acoustic nonlinearity information for diagnosis liver fibrosis. We hypothesize that CNN can extract useful nonlinear information embedded in echo signals to diagnose liver fibrosis instead of computing exact nonlinearity parameter B/A. Then, liver fibrotic stages can be classified by CNN.