The present paper describes the design of pellicle-milk double-layer phantom experiment. Milk solution of 40 different concentrations represents internal information of tissue, 1 to 5 pellicle which covers above the milk solution represents interference information of superficial tissue. The experiment collected 200 scattering spectral data of two positions and took the one single position spectral group as control, and then respectively predicted the milk solution concentration on bottom layer with the ratio of 3:1 through the BP neural network method. The experimental results show that single position scattering spectrum and two-position scattering spectrum both reached more than 90% training fitting rates and prediction accuracy, and the prediction accuracy of two-position scattering spectra is higher, reaching 98.41%. It was verified by the experimental results that scattering spectrum based on photon dissemination path can efficiently predict the milk solution concentration and eliminate the influence of superficial tissue for measurement of internal organization, and considering multi-position in modeling process can improve the accuracy of the prediction. This study validates the feasibility of the method for exploring internal information of tissue without damaging tissue integrity.