A method of conducting sample reflectance testing on a pathological glass substrate and calculating an infrared absorption spectrum

Abstract

The present application relates to the technical field of infrared spectrum testing, and specifically discloses a method for carrying out sample reflection testing on a pathological glass substrate and calculating an infrared absorption spectrum, comprising: S1, placing tissue section samples on an infrared calcium fluoride substrate and a pathological glass substrate respectively for testing, and obtaining a pathological glass substrate spectrum recovery network training data set; S2, training a pathological glass substrate spectrum recovery network constructed based on a deep neural network model by using the obtained pathological glass substrate spectrum recovery network training data set; and S3, obtaining a reflection signal on the pathological glass substrate, and recovering an absorption spectrum by the pathological glass substrate spectrum recovery network. The present application adopts infrared reflection mode to test on a pathological glass substrate sample, mines features, and calculates the infrared absorption spectrum of the sample, and finally can obtain a spectrum compatible with a conventional spectrum analysis method based on pathological glass substrate testing, which is conducive to promoting the technology to the clinic.

Description

Technical Field

[0001] This invention relates to the field of infrared spectroscopy testing technology, and in particular to a method for conducting sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate. Background Technology

[0002] Infrared spectroscopy instruments and analytical methods have made significant progress in recent years, becoming a crucial research field. This technology, based on the absorption of mid-infrared light by various chemical bonds such as SH, CH, NH, and OH, calculates the infrared absorption spectrum of a substance to reflect its characteristic molecular structure, finding wide and diverse applications in the food industry, environmental research, and chemical analysis. In biomedical and clinical research, Fourier transform infrared microscopy has proven highly valuable due to its ability to detect natural tissue biomarkers (collectively known as the disease metabolome) such as proteins, lipids, nucleic acids, and carbohydrates. In many infrared pathology studies, researchers have linked these biomarkers to disease progression, greatly improving our understanding of diseases affecting various organs, including the brain, breast, colon, kidney, liver, lung, ovary, skin, and thyroid. However, several factors currently hinder the clinical application of infrared spectroscopy, one of which is the incompatibility of the specialized substrates required for preparing infrared spectroscopy samples with clinical testing procedures.

[0003] In infrared spectroscopy, a common method for pathological examination, and the calculation of absorption spectra, transmission and reflection tests require special substrates. In transmission mode, infrared light must pass through the sample and then through the substrate before being detected by the detector. Therefore, a substrate with high transmittance in the infrared band is required, commonly calcium fluoride (CaF2) and barium fluoride (BaF2). The drawback of these two substrates is their fragility and high cost. In reflection mode, infrared light passes through the sample, is reflected from the substrate, passes through the sample again, and is finally detected by the infrared detector. Therefore, a substrate with high reflectance in the infrared band is required, commonly using a low-emissivity (Low-e) glass substrate with a metallic coating. However, this coating hinders the transmission of visible light, meaning that biological tissue samples placed on this substrate cannot be examined under a microscope. In traditional pathological examinations, the most commonly used substrate is a 1mm thick transparent glass substrate. Currently, a large number of samples are preserved on glass substrates for microscopic examination and other traditional pathological procedures.

[0004] Studies have shown that during infrared spectroscopy testing of glass, particularly in the fingerprint region (wavenumber range of 1800-1000 cm⁻¹), glass exhibits strong absorption, preventing infrared light from penetrating the substrate and rendering transmission mode unusable. While a weak signal can still be detected in reflection mode, the signal's shape and intensity differ significantly from normally obtained infrared spectra due to the unique optical properties of glass and tissue. This makes it incompatible with well-developed spectral post-processing and analysis algorithms. In this context, recovering the desired transmittance spectrum from the original optical signal becomes a direct and effective method.

[0005] In recent years, deep learning methods and theories have been effectively developed, and deep learning methods are often used in the recovery of 1D and 2D signals. In the development of deep learning, deep neural networks (DNNs) have demonstrated strong feature extraction and feature integration capabilities, and are often used for the conversion between two different modalities of data.

[0006] Therefore, if deep learning methods can be used to perform infrared spectroscopy tests on glass substrates, a large number of previously stored samples can be used for research and testing in infrared spectroscopy pathology, and it can also help to integrate infrared spectroscopy-based pathology tests into clinical pathology examinations. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method for conducting sample reflectance testing and calculating infrared absorption spectra on pathological glass substrates, so as to use a large number of previously stored samples for research and testing in infrared spectroscopy pathology, and at the same time help to integrate infrared spectroscopy-based pathological tests into clinical pathological examinations.

[0008] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0009] A method for performing sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate includes:

[0010] S1. Place tissue slices on infrared calcium fluoride substrate and pathological glass substrate respectively for testing to obtain the training dataset of the pathological glass substrate spectral recovery network.

[0011] S2, Use the obtained pathological glass-substrate spectral recovery network training dataset to train the pathological glass-substrate spectral recovery network based on a deep neural network model;

[0012] S3, the reflected signal is acquired on the pathological glass substrate, and the absorption spectrum is recovered by the pathological glass substrate spectral recovery network.

[0013] Preferably, step S1 includes:

[0014] Tissue section samples were placed on an infrared calcium fluoride substrate to obtain the transmission infrared spectrum, and the Ground Truth absorption spectrum was calculated using negative logarithmic transformation.

[0015] Tissue slide samples are placed on a pathological glass substrate for reflectance testing to obtain reflected infrared spectral signals. These signals are then paired with the Ground Truth absorption spectra obtained at the corresponding positions on an infrared calcium fluoride substrate to form an "input-output" dataset, creating training and testing datasets.

[0016] Preferably, step S1 includes:

[0017] Identical tissue section samples are analyzed using adjacent tissue sections.

[0018] Preferably, the spectrometer used in step S1 is a vacuum cavity infrared spectrometer and an external microscope equipped with an electric translation stage.

[0019] Preferably, the aperture size used in step S1 is 100×100um. 2 The test step size in both the x and y directions is 100um.

[0020] Preferably, step S2 includes:

[0021] The reflectance spectrum signal obtained from the pathological glass substrate is used as the input signal and fed into the pathological glass substrate spectral recovery network.

[0022] The training process compares the absorption spectrum calculated by the spectral restoration network with the absorption spectrum of the Ground Truth obtained by the negative logarithmic transformation of the transmission spectrum of the calcium fluoride substrate sample, calculates the loss function, updates the network parameters through gradient backpropagation, and trains the spectral restoration network for pathological glass substrates.

[0023] The network parameters are continuously updated via gradient backpropagation until the loss function converges, thus obtaining the trained pathological glass substrate spectral recovery network.

[0024] Preferably, step S2 includes:

[0025] The pathological glass substrate spectral recovery network was trained using PyTorch and Python, with a total of 500 training rounds and a learning rate starting from 10%. -3 dropped to 10 -5 The loss function used is as follows:

[0026] L = L MSE +L MAE

[0027]

[0028] In the formula, L MSE The root mean square error between absorption rates; L MAE First derivative The mean absolute error between them; A is the spectral absorbance; A out (v) represents the absorption rate recovered by the comparison network on the wavelength v channel, representing the loss function. target (v) represents the absorption rate of the Ground Truth on the wavelength v channel; V represents the number of wavelengths.

[0029] Preferably, the thickness of the pathological glass substrate is 1 mm.

[0030] Compared with the prior art, the beneficial technical effects of the present invention are:

[0031] This invention can obtain infrared reflectance spectral signals on a pathological glass substrate through a reflectance mode, and the pathological glass substrate used is transparent and without metal coating for clinical pathological examination. This invention can use deep learning methods to recover the infrared absorption spectrum from the reflectance signal of the substrate, and the recovered infrared absorption spectrum can be directly used in subsequent spectral analysis algorithms. Attached Figure Description

[0032] Figure 1 This is a schematic diagram illustrating sample preparation and dataset testing in one embodiment of the present invention:

[0033] Figure 2 This is a schematic diagram of the absorption spectrum input and output design calculation framework for glass substrate testing based on GLASSR-Net in one embodiment of the present invention;

[0034] Figure 3 This is a comparison chart of the results of infrared absorption spectrum recovery calculations in one embodiment of the present invention. Detailed Implementation

[0035] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0036] A method for conducting sample reflectance testing and calculating infrared absorption spectra on pathological glass substrates first obtains infrared signals from the glass substrate using a reflectance test mode. Since the signal shape differs from the transmission signal used in traditional methods, and its intensity is very weak, the transmittance cannot be directly calculated. Therefore, a negative logarithmic transformation is performed to obtain the absorption spectrum, as shown in equation (1):

[0037]

[0038] In the formula, A represents the absorptivity; I0 represents the intensity of the incident light at that wavelength; and I represents the intensity of the emitted light at that wavelength.

[0039] This paper employs deep learning to recover the infrared absorption spectrum obtained from tests on infrared high-transmittance substrates from test signals. Specifically, the deep learning framework primarily utilizes a deep neural network to construct the spectral recovery network for pathological glass substrates.

[0040] The testing method includes the following steps:

[0041] A method for performing sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate includes:

[0042] S1. Place tissue slices on infrared calcium fluoride substrate and pathological glass substrate respectively for testing to obtain the training dataset of the pathological glass substrate spectral recovery network.

[0043] Step S1 includes: preparing tissue section samples, placing adjacent sections on an infrared calcium fluoride substrate, a pathological glass substrate, and another pathological glass substrate, respectively. The first sample placed on the infrared calcium fluoride substrate is used to obtain the transmitted infrared spectrum, and the Ground Truth absorption spectrum is calculated using a negative logarithmic transformation. The second sample placed on the pathological glass substrate is used for reflectance testing to obtain the reflected infrared spectral signal of the sample on the pathological glass substrate, thus forming an "input-output" pair with the Ground Truth absorption spectrum obtained at the corresponding position on the first sample, thereby forming a training and testing dataset. The third sample placed on the pathological glass substrate is used for traditional H&E staining to obtain the pathological characteristics of the sample. The process of preparing this dataset is as follows: Figure 1 Figure (a) shows a schematic diagram of adjacent section preparation; Figure (b) shows a schematic diagram of Tround Truth data acquisition, calculating transmittance by acquiring transmission data on a calcium fluoride substrate; and Figure (c) shows a schematic diagram of data acquisition for testing on a pathological glass substrate, used for data recovery. The aperture size used in the test was 100×100µm. 2 The test step size in both the x and y directions is 100 μm. The spectrometer used is a vacuum cavity infrared spectrometer and an external microscope equipped with an electric translation stage.

[0044] S2, The pathological glass substrate spectral recovery network is trained on a deep neural network model using the obtained pathological glass substrate spectral recovery network training dataset;

[0045] Step S2 includes: training the network using an input-output paired dataset, the network input-output flow as follows: Figure 2The network was trained using PyTorch 2.0.1 and Python 3.10.14. The training consisted of 500 epochs, with a learning rate starting from 10%. -3 dropped to 10 -5 The loss function used is as follows (where the two Loss components are used to recover the shape and control the smoothness of the signal, respectively):

[0046] L = L MSE +L MAE (2)

[0047]

[0048] In the formula, L MSE The root mean square error between absorption rates; L MAE First derivative The mean absolute error between them; A is the spectral absorbance; A out (v) represents the absorption rate recovered by the comparison network on the wavelength v channel, representing the loss function. target (v) represents the absorption rate of the Ground Truth on the wavelength v channel; V represents the number of wavelengths.

[0049] Step S2 includes:

[0050] The reflectance spectrum signal obtained from the pathological glass substrate is used as the input signal and fed into the pathological glass substrate spectral recovery network.

[0051] The training process compares the absorption spectrum calculated by the spectral restoration network with the absorption spectrum of the Ground Truth obtained by the negative logarithmic transformation of the transmission spectrum of the calcium fluoride substrate sample, calculates the loss function, updates the network parameters through gradient backpropagation, and trains the spectral restoration network for pathological glass substrates.

[0052] The network parameters are continuously updated via gradient backpropagation until the loss function converges, thus obtaining the trained pathological glass substrate spectral recovery network.

[0053] S3, the reflected signal is acquired on the pathological glass substrate, and the absorption spectrum is recovered by the pathological glass substrate spectral recovery network.

[0054] The recovered spectra of the test set were compared with the true spectra. The recovered results for different spectral regions were compared with the true results. RMSE is a measure of the difference between signals; the smaller the RMSE value, the closer the two signals are. It can be seen that the deep neural network achieves good recovery results. Figure 3 .

[0055] In one embodiment of the present invention, the thickness of the pathological glass substrate is 1 mm.

[0056] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A method for conducting sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate, characterized in that, include: S1. Place tissue slices on infrared calcium fluoride substrate and pathological glass substrate respectively for testing to obtain the training dataset of the pathological glass substrate spectral recovery network. S2, Use the obtained pathological glass-substrate spectral recovery network training dataset to train the pathological glass-substrate spectral recovery network based on a deep neural network model; S3, the reflected signal is acquired on the pathological glass substrate, and the absorption spectrum is recovered by the pathological glass substrate spectral recovery network.

2. The method for conducting sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate according to claim 1, characterized in that, Step S1 includes: Tissue section samples were placed on an infrared calcium fluoride substrate to obtain the transmission infrared spectrum, and the Ground Truth absorption spectrum was calculated using negative logarithmic transformation. Tissue slide samples are placed on a pathological glass substrate for reflectance testing to obtain reflected infrared spectral signals. These signals are then paired with the Ground Truth absorption spectra obtained at the corresponding positions on an infrared calcium fluoride substrate to form an "input-output" dataset, creating training and testing datasets.

3. The method for conducting sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate according to claim 2, characterized in that, Step S1 includes: Identical tissue section samples are analyzed using adjacent tissue sections.

4. The method for conducting sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate according to claim 1, characterized in that, The spectrometer used in step S1 is a vacuum cavity infrared spectrometer and an external microscope equipped with an electric translation stage.

5. The method for conducting sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate according to claim 4, characterized in that, The aperture size used in step S1 is 100×100um. 2 The test step size in both the x and y directions is 100um.

6. The method for conducting sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate according to claim 1, characterized in that, Step S2 includes: The reflectance spectrum signal obtained from the pathological glass substrate is used as the input signal and fed into the pathological glass substrate spectral recovery network. The training process compares the absorption spectrum calculated by the spectral restoration network with the absorption spectrum of the Ground Truth obtained by the negative logarithmic transformation of the transmission spectrum of the calcium fluoride substrate sample, calculates the loss function, updates the network parameters through gradient backpropagation, and trains the spectral restoration network for pathological glass substrates. The network parameters are continuously updated via gradient backpropagation until the loss function converges, thus obtaining the trained pathological glass substrate spectral recovery network.

7. The method for conducting sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate according to claim 6, characterized in that, Step S2 includes: The pathological glass substrate spectral recovery network was trained using PyTorch and Python, with a total of 500 training rounds and a learning rate starting from 10%. -3 dropped to 10 -5 The loss function used is as follows: L=L MSE +L MAE In the formula, L MSE The root mean square error between absorption rates; L MAE First derivative The mean absolute error between them; A is the absorbance at that wavelength; A out (v) represents the absorption rate recovered by the comparison network on the wavelength v channel, representing the loss function. target (v) represents the absorption rate of the Ground Truth on the wavelength v channel; V represents the number of wavelengths.

8. The method for conducting sample reflectance testing and calculating infrared absorption spectra on a pathological glass substrate according to claim 1, characterized in that: The thickness of the pathological glass substrate is 1 mm.

Images