Detection of heterogeneity on multi-spectral transmission image based on multiple types of pseudo-color maps

Abstract

In transmission imaging, there are many difficulties in identifying heterogeneity on multi-spectral transmission images due to the strong absorption and scattering of biological tissues. This paper combines the multi-wavelength fusion information with the Faster-RCNN model to achieve the heterogeneous detection on the multi-spectral fusion pseudo-color map. Firstly, the original multi-spectral transmission images of phantom are collected on the self-built experimental platform. Then, the gray-scale level of image is improved by modulation and frame accumulation technique. Thereby, fusion pseudo-color maps (fusion pseudo-color maps of original RGB and gradient image) with high quality are obtained. And by comparing the running time, mean average precision (mAP) and root mean square error (RMSE) of Faster-RCNN model, the optimal iteration times and learning rate of model are determined. Finally, the fusion pseudo-color maps used for testing are input into the best Faster-RCNN model to realize the heterogeneous detection. The results show that the Faster-RCNN model established in this paper can effectively identify 2, 4 and 7 different types of heterogeneity on multi-spectral images, respectively. And the best mAP detection results for 2, 4 and 7 heterogeneities are 97.65%, 96.78% and 97.04%, respectively. In conclusion, this paper achieves the detection of heterogeneity on multi-spectral transmission images by fusing high-quality multi-wavelength transmission images and the Faster-RCNN network model, which will promote the clinical application of transmission imaging in early screening of breast cancer.