Three-step one-way model in terahertz biomedical detection

Terahertz technology has broad application prospects in biomedical detection. However, the mixed 17 characteristics of actual samples make the terahertz spectrum complex and difficult to distinguish, 18 and there is no practical terahertz detection method for medical clinics. Here, we propose a three- 19 step one-way terahertz model, presenting a detailed flow analysis of terahertz technology in the 20 biomedical detection of renal fibrosis as an example: 1) biomarker determination : screening 21 disease biomarkers and establishing the terahertz spectrum and concentration gradient; 2) mixture 22 interference removal : clearing the interfering signals in the mixture for the biomarker in the animal 23 model and evaluating and retaining the effective characteristic peaks; and 3) individual difference 24 removal : excluding individual interference differences and confirming the final effective terahertz 25 parameters in the human sample. The root mean square error of our model is three orders of magnitude lower than that of the gold standard, with profound implications for the rapid, accurate and early detection of diseases.


34
The early detection, early diagnosis and early treatment of diseases directly affect the quality of life 35 and survival rate of patients. Pathological diagnosis is the gold standard for the diagnosis of many 36 diseases. The corresponding technologies include light microscopic morphological detection, 37 immunohistochemical enzyme labelling, fluorescence in situ hybridization and gene rearrangement 38 detection [1][2][3][4][5]. Light microscopic morphological detection can judge the nature of the disease by 39 observing the morphological changes in cells but is subjective because this approach is influenced 40 by the observer's experience and intuition. Immunochemical enzyme labelling and fluorescence in 41 situ hybridization (FISH) both use the fluorescent labelling of specific proteins or nucleic acids for 42 localization and quantitative labelling, and gene rearrangement detection technology is used to 43 detect and diagnose diseases by gene sequencing of lesion tissue. However, the sample processing 44 of these techniques is cumbersome and time-consuming and requires large amounts of reagents and 45 dyes, so the samples cannot be reused, and the judgement of results also has subjective interference. 46 Other spectroscopic methods, such as fluorescence spectroscopy and Raman spectroscopy, can also 47 be used for sample detection. Fluorescence spectrometry mainly uses certain fluorescent dyes to 48 form complexes with substances that do not ordinarily emit fluorescence but are induced to emit 49 fluorescence before determination [6][7][8][9]. The procedure is complicated, the fluorescence intensity 50 is affected by the wavelength and temperature of the stimulated luminescence, and the data stability 51 is poor. Raman spectroscopy exposes the sample to a strong laser and collects the scattering signal 52 for material identification [10,11]. However, the high-power laser easily damages the samples, 53 which is not conducive to secondary detection, and the detection error at low concentrations is large, 54 so the method cannot be applied to the early quantitative analysis of diseases. Therefore, there is an 55 urgent need for rapid, label-free, simply operated, low-sample-loss, and low-cost detection 56 technology for the early diagnosis of diseases. 57 Terahertz (THz) waves (0.1-10 THz) lie between the millimetre wave and infrared band, 58 which is of significant importance to the biological sciences due to providing information 59 complementary to traditional spectroscopic measurements of low-frequency bond vibrations, 60 hydrogen bond stretching, and bond torsions in liquids and gases [12]. Therefore, the collective 61 behaviour (vibration and rotation) characteristics of biomolecules make THz spectroscopy a 62 promising sensing modality for clinical diagnosis [13]. Because of these properties and its non-63 destructiveness, accuracy, rapidity and good penetrability [14], THz spectroscopy also has many 64 other potential applications in several research fields, including physics [15,16], biology [17][18][19][20][21], 65 chemistry [22,23], and medicine [24,25]. At present, substance identification with THz technology 66 is mainly based on the fingerprint peaks of compounds. Some studies have indicated that the 67 approach can be combined with chips, algorithms, reagents and other auxiliary means to improve 68 the recognition accuracy [18,[26][27][28][29][30]. However, THz technology has not been effectively applied in 69 actual medical diagnosis or evaluation, because human tissue is a mixture of many substances, and 70 the THz fingerprint spectra of these substances overlap with each other, which interferes with the 71 quantitative analysis of disease markers or key substances. More importantly, there are individual 72 differences between human bodies, and the proportions of various components are continually 73 changing, which leads to a rapid decrease in the effectiveness of auxiliary means such as algorithms. 74 To date, the application of THz technology to biomedical sample detection is still in the process of 75 exploration, and there is no standardized process or model to assist the accurate diagnosis of 76 diseases. 77 In this paper, we propose a three-step one-way THz model to analyse the application steps and 78 processes of using THz spectroscopy to detect real diseases. We hope that we can apply this model 79 to improve the application of THz technology in research on related diseases and to realize the rapid 80 and effective detection of pathological specimens of various diseases by using THz technology. 81 The THz three-step one-way model is divided into three steps, namely, biomarker 82 determination, mixture interference removal, and individual difference removal. In the first step, 83 we need to screen and determine the biomarkers or key substances (of which there can be multiple), 84 test the THz fingerprint spectrum, and establish the gradient relationship between the biomarker 85 concentration and THz parameters to lay the foundation for subsequent calibration of the disease 86 pathological stage. Second, to verify the possibility of applying THz technology to recognize 87 biomarkers in biological tissues (mixtures), we initially used animal models to predict and quantify 88 these markers in accordance with scientific research ethics. It is possible to analyse whether the 89 biomarkers of rat pathological samples can be identified by observing the pathological samples of 90 rats. The third step is mainly to determine whether the individual differences in the human body 91 will affect the qualitative identification and quantitative analysis of the biomarkers by THz 92 spectroscopy and to verify the validity and accuracy of the biomarker parameters. The above three 93 steps can be implemented only in one direction; otherwise, it is easy to reach a misdiagnosis or 94 incorrect stage identification due to individual differences or mixture identification errors. 95 To assess the feasibility of the THz three-step one-way model, we used renal fibrosis, a 96 common chronic disease, as an example. Renal fibrosis is a common pathological manifestation of 97 end-stage renal disease; its main pathological features are glomerulosclerosis, tubular atrophy and 98 increased extracellular matrix deposition [31]. Studies have shown that the assessment of renal 99 fibrosis can guide the treatment and prognosis of autologous kidneys [32] and transplanted kidneys 100 [33]. There are three methods for the diagnosis of fibrosis: histological evaluation, imaging 101 evaluation and biomarker detection. 1) Histology is the gold standard for the diagnosis of fibrosis 102 and can directly observe the degree of fibrosis, but its identification depends on the subjective 103 judgement of pathologists, and there are individual differences. In addition, in some cases, there are 104 sampling errors in the tissues obtained from core biopsy [34]. Moreover, there are time limitations 105 in histopathological detection: sample preparation takes a long time (approximately 15 hours), 106 preventing rapid determination of the degree of fibrosis. 2) Imaging evaluation methods, including 107 ultrasound and functional magnetic resonance imaging (MRI), are new noninvasive techniques. 108 Renal fibrosis is reflected by observing the renal elasticity, oxygen content and blood perfusion. 109 However, the identification of renal fibrosis is easily affected by many factors, such as blood 110 pressure, body weight, respiratory movement and differences in subjective judgement among 111 observers, resulting in low detection accuracy [35].
3) The use of biomarkers in haematuria as a 112 noninvasive detection method is expected to be usable to monitor the progression of renal fibrosis 113 [36]. However, some promising biomarkers, such as microRNAs [37], still suffer from instability 114 or lack of regularity as the disease progresses. Therefore, we expect to solve these problems through 115 the THz three-step one-way model.  119 The experimental setup used here is THz time-domain spectroscopy. In our experiments, we used 120 an 800 nm femtosecond laser with a pulse duration of 100 fs, a repetition rate of 76 MHz, and an 121 average power of 150 mW. The emitted laser beam was split into a pump beam and probe beam by 122 a 50/50 beam splitter. A pump beam modulated by an optical chopper was focused on a gallium 123 arsenide (GaAs) photoconductive emitter of THz waves. The diverging THz beam was collected 124 and focused by four off-axis paraboloid mirrors to pass though samples, and then, the probe beam was used to detect the THz wave by photoconductive antennas. For our experimental system, the 126 effective bandwidth for the measured signals ranged from 0.2 to 2.0 THz, the spectrum resolution 127 was ~15 GHz, and the signal-to-noise ratio (SNR) was greater than 1000:1. All the spectra were 128 averaged 256 times to ensure a high SNR. The sample absorbance α(ω) was calculated by using the 129 following equation: α(ω)=log (Iref(ω)/Isam(ω))/d, where d=0.1 mm is the thickness of the sample, 130 Isam(ω)= Esam(ω)×Esam(ω) * is the power spectrum of the sample, and Iref(ω)= Eref(ω)× Eref(ω) * is the 131 power spectrum of the reference signal. high-density polyethylene windows is a commercial product designed for optical spectroscopy. The 142 cryostat temperature can continuously vary between 77 K and 500 K, with an accuracy of ±0.5 K. 143 In our experiments, the cryostat was placed at the THz radiation focal point between the off-axis 144 parabolic mirrors of the THz-TDS system. The sample was placed inside in the cryostat, and THz 145 radiation was passed through the sample to obtain the fingerprint spectrum. of 20 male SD rats were randomly divided into the following four groups: the sham surgery group 153 (sham, n=5), the UUO d3 group (d3, n=5), the UUO d5 group (d5, n=5), the UUO d7 group (d7, 154 n=5), and the UUO d14 group (d14, n=5). After anaesthesia by using 3% intraperitoneal 155 pentobarbital (2 ml/kg), the UUO groups received left ureteral ligation in the sham group, and the 156 left ureter was separated, but not ligated. On days 3, 7, and 14 after surgery, the rats were sacrificed 157 using euthanasia. Their left kidneys were removed and homogenized and then placed into sterile 158 tubes that were labelled. All sterile tubes were kept in a 200 K freezer. Then, rat left kidney 159 homogenates were carefully mounted between two flat quartz plates with a groove thickness of 480 160 μm.

161
Masson staining 162 After the kidneys were harvested, they were sliced axially into 3-mm-thick sections, fixed in 10% 163 buffered formalin, embedded in paraffin, and deparaffinized by submersing the slides in 4 changes 164 of xylene for 2 minutes each, then dehydrated in 2 changes of absolute alcohol for 2 minutes, 1 165 change of 95% alcohol for 2 minutes, and 1 change of 70% alcohol for 2 minutes. The slides were 166 then mordanted by submersion in Bouin's fixative for 1 hour at 50-60°C, washed in gently running 167 tap water until the yellow colour completely disappeared, rinsed in distilled 3 times, stained by 168 submersion in Weigert's iron haematoxylin for 10 minutes, rinsed in warm tap water for 15 minutes, 169 stained by submersion in Biebrich scarlet-acid fuchsin for 15 minutes, immersed in 170 phosphomolybdic-phosphotungstic acid for 15 minutes, and immersed in Aniline blue solution for 171 15 minutes. Next, the slides were rinsed with distilled water 3 times, immersed in 1% acetic water 172 for 1 to 3 minutes, rinsed in distilled water 3 times, dehydrated by submersion in 4 absolute alcohol 173 changes for 2 minutes each, and cleared by submersion in 4 xylene changes for 2 minutes each. 174 Finally, the coverslip was mounted in resinous mounting media. Alkaline hydrolysis (AH) 182 First, kidney tissue (40 mg wet weight) was sufficiently mixed with the digest (1 mL). After the 183 test tube was covered by a stopper, the mixture was heated at 96°C for 20 minutes, and the pH was 184 adjusted to 6.0. Second, distilled water was added to the test tube until the total volume reached 10 185 mL. Next, diluted lysate (4 mL) and quantum active carbon (30 mg) were added to the test tube and 186 mixed sufficiently. The mixture was then centrifuged at 3500 rpm for 10 minutes. The supernatant 187 (1 mL) was transferred into a clean tube. The remainder of the experimental process is described in 188  196 Kidney tissue (10 mg) was cut into slices and then transferred into a screw vial containing 6 mol/L 197 HCl (1 mL). After the vial cap was screwed on tightly, the mixture was heated for 6 hours at 110°C. 198 The hydrolysed mixture was then centrifuged at 12000 rpm for 10 minutes at 4°C.     Table 3.   259   Table 3. Calculated absorption peaks of L-hydroxyproline and vibration model analysis.

1.64 THz
The combination of the right-left vibration of the right molecule and the strong contractions of the left molecule [see Fig. 1f].

261
After confirming that the absorption peaks were located within the THz spectral window, we  Table 4.
274 Similar to the classical case, peak 3 undergoes a redshift as the temperature increases. This  We analysed the results of the LC-MS, AH and THz tests by using the one-way analysis of 399 variance (ANOVA) method, as shown in Fig. 3d&3f&3h. It can be clearly seen that for both the 400 LC-MS and AH methods, the P values were all greater than 0.05 (PAH=0.6293, PMS=0.7286) in the 401 second stage (Fig. 3d&1f), which indicated that the early stage of fibrosis could not be effectively 402 distinguished. By comparison, THz detection showed good differentiation (Fig. 3h). For the third 403 and fourth stages, all three methods effectively identified fibrosis, but the THz method had a smaller 404 error and higher discrimination. Therefore, the accuracy of early diagnosis for the THz method is 405 much higher than that of the other two conventional methods. According to Masson staining, 14 renal tissues were divided into a normal group (n = 4) and 426 a fibrosis group (n = 10). The average age of the normal group was 50.00 ± 9.41 years (mean eGFR 427 66.75 ± 6.28 mL/min), and the fibrotic area was 4.00 ± 0.82%; the average age of the fibrosis group 428 was 57.00 ± 3.25 years, the average eGFR was 65.36 ± 8.30 mL/min, and the fibrotic area was 429 30.40 ± 8.50%. Fig. 4a shows a pathological image of normal human kidney tissue. The glomerular 430 capillary loop is well opened, and the renal tubules are back-to-back. No obvious tubular atrophy, 431 interstitial fibrosis or inflammatory cell infiltration was found. Fig. 4b presents a pathological image 432 of a human fibrotic kidney with obvious glomerulosclerosis, tubular atrophy with interstitial 433 fibrosis and inflammatory cell infiltration. 434 The THz absorption spectra of human samples are shown in Fig. 4c. The spectra of human 435 kidney tissue changed again compared to those of the rat model. The central frequency of the first 436 characteristic peak at 0.69 THz in the rat sample redshifted slightly to 0.64 THz, and the central 437 frequency of the characteristic peak at 1.02 THz did not change; the biomarker-unrelated peaks at 438 0.15, 0.41, 1.27, and 1.63 THz in Fig. 3b no longer exhibit obvious or regular changes in Fig. 4c. 439 We speculate that this is because the rats in the UUO model were fed uniformly and injected 440 quantitatively. However, the individual differences among human bodies (the proportion and 441 content of each component varies with individual age, sex, physique, nutrition, exercise and other 442 factors) will change the centre frequency of the overlapping spectrum, which causes the spectrum 443 to no longer have obvious regular changes. However, the content of hydroxyproline, a common 444 characteristic of renal fibrosis patients, increased significantly, making its characteristic peak still 445 more significant and able to be effectively identified. 446 As in UUO, we integrated the area under the peak of the two characteristic peaks at 0.64 and 447 1.02 THz, corresponding to the frequency range of 0.50-0.86 THz and 0.86-1.19 THz, respectively, 448 as shown in the red box in Fig. 4c. The relevant data and AH test results were divided into two 449 groups, normal and pathological, as shown in Fig. 4d and 4e. The results showed that the content 450 of hydroxyproline and the area under the peaks by THz-TDS in renal fibrotic and normal tissues 451 were significantly increased, and there was a significant difference between the two groups by THz 452 intensity and THz peak area assessment. Considering these results and the rapidity, simple operation 453 and losslessness of the THz method, we conclude that the overall effect of the THz method is much 454 better than that of the AH method. In summary, we provide a "THz three-step one-way" model for disease detection, which details the 484 process of THz dynamic monitoring of disease evolution and the method of eliminating 485 interference. Compared with AH and LC-MS, THz-TDS can improve the accuracy, sensitivity and 486 detection speed of diagnosis. Its high sensitivity, high accuracy and high speed highlight its 487 potential in the early diagnosis, staging evaluation and disease monitoring of biological diseases. 488 In the future, this kind of dynamic monitoring is expected to achieve real-time imaging during 489 operation, which can greatly improve the accuracy of surgery.