Detection of heterogeneity in multi-spectral transmission image based on spatial pyramid matching model and deep learning. (November 2020)

• Record Type: Journal Article
• Authors: Liu, Fulong; Li, Gang; Yang, Shuqiang; Yan, Wenjuan; He, Guoquan; Lin, Ling
• Is Part Of: Optics and lasers in engineering. Volume 134(2020)
• Publication Date: 2020-11
• Subjects: Multi-spectral transmission image -- Modulation-demodulation-frame accumulation technique (MDFAT) -- Spatial pyramid matching (SPM) model -- Faster-RCNN -- Single Shot Multi-Box Detector (SSD) -- Heterogeneous detection; Lasers in engineering -- Periodicals; Optical measurements -- Periodicals; Optics -- Periodicals; Lasers en ingénierie -- Périodiques; Mesures optiques -- Périodiques; Optique -- Périodiques; 621.36605
• DOI: 10.1016/j.optlaseng.2020.106272
• Languages: English
• ISSNs: 0143-8166

Abstract

Highlights: The recognition of heterogeneities has many difficulties due to the obvious scattering effect of the light source in the transmission process of biological tissues. The quality and clarity of multi-spectral transmission image is improved by the modulation and demodulation-frame accumulation technique (MDFAT) and spatial pyramid matching (SPM) model. Combined with the feature information of multi-spectral transmission images, the heterogeneities of multi-spectral images are effectively recognized in the deep learning network (Faster-RCNN and SSD). Multi-spectral transmission imaging provides a possibility for the early detection of breast cancer. Abstract: The absorption and scattering effects of light source during the transmission of biological tissues make it difficult to identify heterogeneity in multi-spectral images. This paper firstly proposes a combination method of modulation-demodulation-frame accumulation technique (MDFAT), spatial pyramid matching (SPM) model and deep learning to realize heterogeneous detection in multi-spectral images. Firstly, the acquisition experiment of phantom image is designed. Then, the high-quality multi-spectral images are obtained by the MDFAT and SPM model. Finally, the pseudo-color maps of high-quality multi-spectral images fusion are served as the input of Faster-RCNN and Single Shot Multi-Box Detector (SSD) network models to realize heterogeneous detection. The results show that Faster-RCNN and SSD both have good detection results. Among them, Faster-RCNN model has the best detection effect on the images containing three types of heterogeneity, and the mean average precision (mAP) reaches 93.91%. SSD model has the most ideal detection effect for the images containing two and five types of heterogeneity, with mAP reaching 94.16% and 94.78% respectively. In conclusion, this paper has verified the feasibility of detecting heterogeneities in multi-spectral images through deep learning network (Faster-RCNN and SSD), which will promote the clinical application of multi-spectral transmission imaging in early screening of breast tumors.