Yuan Zhao's research while affiliated with Nanjing University of Aeronautics & Astronautics and other places

When a convolutional neural network (CNN) model was built, the size and resolution of its input data were fixed. However, Raman spectra collected by different Raman spectrometers usually had different length, intensity range, and wavenumber interval between two adjacent data points, which made the existing CNN model difficult to be applied to a new Raman spectral data set. Therefore, this paper proposed a polynomial reconstruction algorithm as pretreatment method to obtain reconstructed spectra that would be imported into CNN model with consistent length, intensity range, and wavenumber interval. To test the effectiveness of this method, a big data set with 2563 Raman spectra of 831 minerals and synthetic organic pigments samples was constructed from the RRUFF and SOP database to pretrain a one‐dimensional CNN (1D‐CNN) model. The pretraining results showed that polynomial reconstruction algorithm used as pretreatment method was better than SG smoothing combined spline interpolation algorithm. Then two data sets were collected by different Raman spectrometers for evaluating the transfer learning performance of the trained 1D‐CNN model. Both data sets contained 390 Raman spectra from the same 39 samples of inorganic salts, organic compounds, and amino acids. One was used as calibration data to retrain the 1D‐CNN model, while the other was used as test. Based on data augmentation and 75% calibration data for retraining, the transfer learning performances of 1D‐CNN model were clearly shown in the excellent identification accuracies of 99.58%, 99.32%, and 97.69% for training, validation, and test sets, respectively, which were better than those of K‐nearest neighbor classifier. This paper provides a significant way for the wide application of CNN model in Raman spectroscopy with much more advantages in simplicity and rapidity.

A home-made small-sized Raman spectrometer combined with machine learning algorithms was used to study and identify healthy and multi-period osteoarthritis (OA) canine knee joints. Nine canines were equally divided into three groups according to the post-operative (OA modeling) time of 2-month, 3-month and 7-month. Other two normal canines were used as control. It was found that the degeneration degree of cartilage was positively correlated with post-operative time by doing anatomical analysis. The mixed Raman spectra of cartilage and subchondral bone were collected and analyzed, which reveals subchondral bone demineralization and carbonate ion substituting into the apatite mineral during OA. Raman spectra combined with principal component analysis (PCA) further disclosed that collagen matrix became unordered, both content ratios of amide I/matrix and phenylalanine/matrix in OA cartilage and subchondral bone increased. Based on the PCA getting five principal components, all groups were effectively discriminated by Fisher discriminant analysis (FDA) with high accuracy of 91.07% for the validation set, as well as 95.45% for the test set. It suggests that Raman spectroscopy combined with machine learning is capable to become an effective tool to achieve in situ identification of multi-period OA with high accuracy and preclinical significance.

Deep learning is usually combined with a single detection technique in the field of disease diagnosis. This study focused on simultaneously combining deep learning with multiple detection technologies, fluorescence imaging and Raman spectroscopy, for breast cancer diagnosis. A number of fluorescence images and Raman spectra were collected from breast tissue sections of 14 patients. Pseudo-color enhancement algorithm and a convolutional neural network were applied to the fluorescence image processing, so that the discriminant accuracy of test sets, 88.61%, was obtained. Two different BP-neural networks were applied to the Raman spectra that mainly comprised collagen and lipid, so that the discriminant accuracy of 95.33% and 98.67% of test sets were gotten, respectively. Then the discriminant results of fluorescence images and Raman spectra were counted and arranged into a characteristic variable matrix to predict the breast tissue samples with partial least squares (PLS) algorithm. As a result, the predictions of all samples are correct, with minor error of predictive value. This study proves that deep learning algorithms can be applied into multiple diagnostic optics/spectroscopy techniques simultaneously to improve the accuracy in disease diagnosis.

Submillimetric hollow fiber (HOF-) attenuated total reflection (ATR-) Fourier transform infrared spectroscopy (FTIRS) technique is very suitable for in situ detection of biological samples. HOF-ATR-FTIRS analysis discloses that the significant changes in collagen fiber and proteoglycan between healthy and osteoarthritic (OA) canine specimens. The technique combined with principal component analysis (PCA) and Fisher’s discriminant analysis (FDA) was used to further identify in situ healthy and OA cartilage specimens. The sensitivity of 100% and specificity of 93.6% were obtained for prediction group after all the cases were correctly identified for initial sample, and cross validation was identified with 100% accuracy. The HOF-ATR-FTIRs technique is convenient for operation and suitable for micro-area in situ diagnosis, while the FDA method is fast and accurate. The combination of HOF-ATR-FTIR technique and FDA method can fast realize the micro-area detection and classification of in situ samples with good accuracy, which has potential ability to provide a practical clinical diagnosis of OA.

To accurately investigate in situ breast cancer would be very significant for real-time information and in situ diagnosis. In this in situ study, home-made hollow optical fiber attenuated total reflection (HOF-ATR) probe was integrated into Fourier transform infrared (FTIR) spectroscopic system to perform breast cancer research at molecular level. Based on the FTIR spectral analysis on band shifts and absorbance ratios, it's disclosed that the molecular structure, conformation and content of main components change with cancerization of breast tissue. Fisher's discriminant analysis on HOF-ATR-FTIR spectra was applied to identify the healthy and cancerous breast tissues for the first time. The identification accuracy was 96.67% for training group and 93.33% for cross-validation, respectively, as well as 95% for the prediction group. This paper provides much in situ information of tissue cancerization at molecular level, which can be used as fingerprint biomarkers of tissue cancerization for in situ diagnosis. HOF-ATR-FTIR spectroscopy with discriminant analysis has potential to be an effective and promising method in in situ biomedical research and monitoring of breast cancer.

Fourier transform infrared imaging (FTIRI) can be used to obtain the composition and structure information of sample. Here, FTIRI combined with spectral polarization analysis method was applied to investigate the fine anisotropy of bovine nasal cartilage (BNC). The upper BNC tissue was sliced into a three-dimensional (3D) block with three planes (XY, YZ, and XZ) parallel to horizontal section, forward section, and lateral section, respectively. The anisotropy of collagen fiber in BNC was represented by the absorbance of amide II (1590–1500 cm⁻¹) at different polarization directions. It was found that collagen fiber showed little anisotropy in plane XY, XZ, and along the direction Z in plane YZ. It was more important that collagen fiber showed strong anisotropy along direction Y in plane YZ (transverse axis) of BNC, possibly including arched or wavy fiber orientation even a mixture of both in nasal septum top end. Two anisotropic deflections ranging from 600 to 930 μm and from 2680 to 2980 μm were quantitatively calculated. This study is of important significance for further understanding the physiological structure of nasal septum and provides remarkable experimental support for being a good transplant material in cartilage reshaping studies.

A novel attenuated total reflection-mid infrared-hollow optical fiber (ATR-MIR-HOF) probe was home-built, which mainly consists of two HOFs and a newly designed conical frustum of ZnSe crystal. Compared to conventional Fourier transform infrared (FTIR) and ATR-FTIR imaging technique, ATR-HOF-FTIR could obtain reliable ATR IR spectra of samples without any sample pretreatment. ATR-HOF-FTIR detection of articular cartilage (AC) further demonstrates that the ATR-MIR-HOF probe is capable to easily detect in situ tiny target area and obtain effective spectral information on the changes in main constituents of AC with excellent repeatability. The probe is suitable for submillimetric detection in biochemical, biomedical areas and is potential for clinical diagnosis of osteoarthritis at different stages.

Osteoarthritis (OA) is not only related to the degradation of articular cartilage, but also possibly to the changes of subchondral bone. The purpose of this study was to assess whether specific differences could be resolved from bone composition, as also contributed to OA. These differences were assessed by using Fourier transform infrared spectroscopy (FTIRS). The main parameters including mineral content, carbonate content, crystallinity, collagen cross-linking ratio (XLR) and acid phosphate content were represented with characteristic peak integration. It was found that mineral and carbonate content varied significantly with depths at different OA stages. Mineral content increased with depth in healthy samples, while carbonate content showed opposite trend. The mineral content reduced obviously with OA duration, which was different with carbonate decreasing only at early stage of OA. In addition, the content of acid phosphate, collagen maturity (XLR) and crystallinity slight varied with the OA aggravation. Therefore, the changes in subchondral bone were significantly associated with cartilage degeneration and OA, the associated parameters should be targeted for OA therapies.