Training Universal Deep-Learning Networks for Electromagnetic Medical Imaging Using a Large Database of Randomized Objects

Abstract

Deep learning has become a powerful tool for solving inverse problems in electromagnetic medical imaging. However, contemporary deep-learning-based approaches are susceptible to inaccuracies stemming from inadequate training datasets, primarily consisting of signals generated from simplified and homogeneous imaging scenarios. This paper introduces a novel methodology to construct an expansive and diverse database encompassing domains featuring randomly shaped structures with electrical properties representative of healthy and abnormal tissues. The core objective of this database is to enable the training of universal deep-learning techniques for permittivity profile reconstruction in complex electromagnetic medical imaging domains. The constructed database contains 25,000 unique objects created by superimposing from 6 to 24 randomly sized ellipses and polygons with varying electrical attributes. Introducing randomness in the database enhances training, allowing the neural network to achieve universality while reducing the risk of overfitting. The representative signals in the database are generated using an array of antennas that irradiate the imaging domain and capture scattered signals. A custom-designed U-net is trained by using those signals to generate the permittivity profile of the defined imaging domain. To assess the database and confirm the universality of the trained network, three distinct testing datasets with diverse objects are imaged using the designed U-net. Quantitative assessments of the generated images show promising results, with structural similarity scores consistently exceeding 0.84, normalized root mean square errors remaining below 14%, and peak signal-to-noise ratios exceeding 33 dB. These results demonstrate the practicality of the constructed database for training deep learning networks that have generalization capabilities in solving inverse problems in medical imaging without the need for additional physical assistant algorithms.

Keywords: antenna sensing; database; deep learning; electromagnetic imaging.

Figure 1

The workflow for building an imaging domain with mixed geometric objects and collecting EM data.

Figure 2

Database composition and geometry settings in the imaging domain.

Figure 3

The procedure of DL-based permittivity profile reconstruction and the strategy for training and testing.

Figure 4

The architecture of the U-net is used in DL-based permittivity reconstruction.

Figure 5

These are representative results of the proposed database-trained network in the Data A testing. The ground truths of scatterers and the reconstructed permittivity distributions are shown in the first and second rows, respectively. The red dotted box represents the expected regions with high permittivity values.

Figure 6

Representative results of the proposed database-trained network in the Data B testing. The ground truths of scatterers and the reconstructed permittivity distributions are shown in the first and second rows, respectively. The red dotted box represents the expected regions with high permittivity values.

Figure 7

Representative results of the proposed database-trained network in the MNIST data testing. The ground truths of scatterers (1, 3 rows) and the reconstructed permittivity distributions (2, 4 rows) are shown, respectively. The red dotted box represents the expected digits imaging areas with high permittivity values.

Figure 8

Representative results of MNIST-trained network in all the test datasets. The ground truths of scatterers (1, 3, and 5 rows) and the reconstructed dielectric distributions (2, 4, and 6 rows) are shown, respectively. The red dotted box represents the expected imaging areas with high permittivity values.

Figure 9

The evaluation histograms of Data A trained network using SSIM, NRMSE, and PSNR for all reconstruction profiles in both the test datasets. (Apart from the blue, pink and yellow bars, the bars in other colors are the overlapping parts of these three bars.)

Figure 10

The evaluation histograms of the Data MNIST-trained network using SSIM, NRMSE, and PSNR for all reconstruction profiles in all the test datasets. (Apart from the blue, pink and yellow bars, the bars in other colors are the overlapping parts of these three bars).