Medical Image Analysis
Medical Image Analysis provides a forum for the dissemination of new research results in the field of medical and biological image analysis, with special emphasis on efforts related to the applications of computer vision, virtual reality and robotics to biomedical imaging problems. Many more exciting developments are on the horizon. These include graph neural networks, knowledge graphs, and semisupervised and unsupervised learning, just to name a few. Successful applications of AI-based medical imaging techniques will be realized in the next 5–10 years, generating a huge impact on health care in general.
Methods
An overwhelming majority of researchers are working on deep imaging now worldwide. The efficacy and efficiency of deep learning is undeniable in the medical imaging field, as evidenced by a sufficiently large number of independent studies across modalities and application areas. Along with the impressive results, related practice issues become pressing, involving interpretability, interoperability, robustness, and optimality, in a continuous learning environment. Deep learning methods have become dominant contenders both in number of submissions and performance.
GAN-based segmentation methods mainly use GAN/cGAN/DCGAN in addition to pixel/voxel-wise optimization loss functions. CycleGAN and reconstruction loss strategies are also proposed to consider non-RoI features for more precise segmentation. In the reviewed methods, U-Net and ResNet—due to providing general identification features—are the most popular segmentation networks for the generator architecture.These GAN models can be used in medical image analysis of the spine. Moreover, GANs made a significant enhancement in microscopic image segmentation.
AI algorithms for medical image analysis can be divided into several main groups.
| Type of algorithm |
Examples |
Type of task |
Dataset size |
Interpretability |
Overfitting |
Training time |
| Linear classifier |
Linear regression, logistic regression, support vector machines (SVM) with linear kernels |
Classification |
Works well with smaller datasets |
Easily interpreted |
Not prone to overfitting |
Fast |
| Non-linear classifier |
SVM using non-linear kernels |
Classification and Regression (in certain cases) |
Not suitable for very large datasets |
Difficult to interpret |
Medium |
Slow |
| Decision trees |
Regular or with boosting (XGBoost) and bagging (RandomForest) |
Classification and regression |
Works well on large datasets |
Easy to interpret |
Prone to overfitting |
Fast |
| Neural Networks |
Convolutional neural networks (CNN), Deep neural networks (DNN) Recurrent neural networks (RNN) |
Classification and regression |
Requires large datasets |
Difficult to interpreted, "black box" nature |
Prone to overfitting, increases with model complexity |
Slow |
Table 1 Overview of AI algorithms and their attributes.
Our Medical Image Analysis Services
CD ComputaBio provides medical image analysis software for hospital and practices worldwide. In healthcare, computer vision is used as part of computer-aided diagnosis software helping doctors to detect abnormal signs in both 3D and plain images and videos. While not being able to replace human clinicians, computer vision applications can become their reliable assistants and thus reducing human error in radiology.
