Microwave image reconstruction based on a deep learning method is investigated in this article. The neural network is capable of converting measured microwave signals acquired from a 24 × 24 antenna array at 4 GHz into a 128 × 128 image. To reduce the training difficulty, we first developed an autoencoder by which high-resolution images (128 × 128) were represented with 256 × 1 vectors; then we developed the second neural network which aimed to map microwave signals to the compressed features (256 × 1 vector). Two neural networks can be combined to a full network to make reconstructions, when both are successfully developed. The present two-stage training method reduces the difficulty in training deep learning networks (DLNs) for inverse reconstruction. The developed neural network is validated by simulation examples and experimental data with objects in different shapes/sizes, placed in different locations, and with dielectric constant ranging from 2 to 6. Comparisons between the imaging results achieved by the present method and two conventional approaches: distorted Born iterative method (DBIM) and phase confocal method (PCM) are also provided.
12 May 2020 in IEEE Transactions on Radiation and Plasma Medical Sciences
A neural network designed specifically for single photon emission computed tomography (SPECT) image reconstruction was developed. The network reconstructed activity images from SPECT projection data directly. Training was performed through a corpus of training data, including that derived from digital phantoms generated from custom software and the corresponding projection data obtained from simulation. When using the network to reconstruct images, input projection data were initially fed to two fully connected (FC) layers to perform a basic reconstruction. Then, the output of the FC layers and an attenuation map were delivered to five convolutional layers for signal-decay compensation and image optimization. To validate the system, data not used in training, simulated data from the Zubal human brain phantom, and clinical patient data were used to test reconstruction performance. Reconstructed images from the developed network proved closer to the truth with higher resolution and quantitative accuracy than those from conventional OS-EM reconstruction. To understand better the operation of the network for reconstruction, intermediate results from hidden layers were investigated for each step of the processing. The network system was also retrained with noisy projection data and compared with that developed with noise-free data. The retrained network proved even more robust after having learned to filter noise. Finally, we showed that the network still provided sharp images when using reduced view projection data (retrained with reduced view data).
2 January 2017 in IEEE Transactions on Microwave Theory and Techniques
A new fast imaging algorithm only using the phase information of electromagnetic fields for near-field microwave imaging is presented. This algorithm first compensates the phase shift as the wave back-propagates to the source, and then calculates the variance of phase of the shifted signals in terms of unit vector. This process is repeated at all focal points in the interested domain to form an image. The algorithm was validated by simulation and experiments; acceptable results were seen in both methods. To carry out the experimental test, we developed a fully automated microwave detection system containing a twin-antenna system whose movement can be controlled to form a synthetic aperture. Experiments were fully controlled by our self-developed software supporting remote operations. Although only 2-D imaging tests are discussed in this paper, the algorithm and the measurement system are able to fulfill 3-D rapid microwave scans.
25 January 2019 in IEEE Transactions on Microwave Theory and Techniques
A deep learning (DL) method is applied to modeling electromagnetic (EM) scattering for microwave breast imaging (MBI). The neural network (NN) accepts 2-D dielectric breast maps at 3 GHz and produces scattered-field data on an antenna array composed of 24 transmitters and 24 receivers. The NN was trained by 18 000 synthetic digital breast phantoms generated by generative adversarial network (GAN), and the scattered-field data pre-calculated by method of moments (MOM). Validation was performed by comparing the 2000 NN-produced datasets isolated from the training data with the data computed by MOM. Finally, data generated by NN and MOM were used for image reconstruction. The reconstruction demonstrated that errors caused by NN would not significantly affect the image result. But, the computational speed of NN was nearly 10 $^{4}$ times faster than the MOM, indicating that DL has the potential to be considered as a fast tool for EM scattering computation.
Machine learning (ML) techniques have attracted many microwave researchers and engineers for their potential to improve performance in microwave- and millimeter-wave-based medical applications. This paper reviews ML algorithms, data acquisition, training techniques, and applications that have emerged in recent years. It also reviews state-of-the-art ML techniques applied for the detection of various organ diseases with microwave signals, achieving more successful results than using traditional methods alone, such as a higher diagnosis accuracy or spatial resolution and significantly improved efficiency. Challenges and the outlook of using ML in future microwave medical applications are also discussed.