Improving heterogeneous classification accuracy based on the MDFAT and the combination feature information of multi-spectral transmission images

Abstract

Multi-spectral transmission image provides a possibility for the early detection of breast cancer. However, in the process of obtaining multi-spectral transmission images, it is difficult to identify the heterogeneity in images due to the image blur caused by the scattering effect of the light source in the biological tissue and the weak transmission signal. This paper proposes a combination method based on the modulation-demodulation-frame accumulation technique (MDFAT) and the combination feature information of multi-spectral transmission images to improve the accuracy of heterogeneous classification. Firstly, the acquisition experiment of phantom multi-spectral transmission images is designed. Then, the high-resolution image is obtained by the MDFAT, and the 14-dimensional feature information of the heterogeneous region is extracted from the images before and after processing. The combination feature information of wavelengths is arranged in order of blue light, green light, near-infrared light and red light. Finally, Random Forest (RF) is used to classify the heterogeneities in the transmission image. The results show that the quality of multi-spectral transmission image is significantly improved after the processing of MDFAT, and the gray level of image is also obviously increased, so that more abundant feature information of heterogeneous region can be obtained. And the overall classification accuracy of RF model established after image preprocessing has been significantly improved. Among them, the 3-wavelength combination model has the best classification effect and the best robustness, followed by the 4-wavelength combination model. The classification accuracy of single-wavelength model is low, but it is also greatly improved compared with that before image preprocessing. In conclusion, this paper improves the image quality by the MDFAT, and the classification accuracy of heterogeneities is significantly improved by combining the feature information of multi-spectral transmission images, which promotes the potential application of multi-spectral transmission imaging in early breast cancer detection.