Machine learning augmented near-infrared spectroscopy: In vivo follow-up of cartilage defects

Infrared Spectroscopy Can Differentiate Between Cartilage Injury Models: Implication for Assessment of Cartilage Integrity

In order to improve the ability of clinical diagnosis to differentiate articular cartilage (AC) injury of different origins, this study explores the sensitivity of mid-infrared (MIR) spectroscopy for detecting structural, compositional, and functional changes in AC resulting from two injury types. Three grooves (two in parallel in the palmar-dorsal direction and one in the mediolateral direction) were made via arthrotomy in the AC of the radial facet of the third carpal bone (middle carpal joint) and of the intermediate carpal bone (the radiocarpal joint) of nine healthy adult female Shetland ponies (age = 6.8 ± 2.6 years; range 4-13 years) using blunt and sharp tools. The defects were randomly assigned to each of the two joints. Ponies underwent a 3-week box rest followed by 8 weeks of treadmill training and 26 weeks of free pasture exercise before being euthanized for osteochondral sample collection. The osteochondral samples underwent biomechanical indentation testing, followed by MIR spectroscopic assessment. Digital densitometry was conducted afterward to estimate the tissue's proteoglycan (PG) content. Subsequently, machine learning models were developed to classify the samples to estimate their biomechanical properties and PG content based on the MIR spectra according to injury type. Results show that MIR is able to discriminate healthy from injured AC (91%) and between injury types (88%). The method can also estimate AC properties with relatively low error (thickness = 12.7% mm, equilibrium modulus = 10.7% MPa, instantaneous modulus = 11.8% MPa). These findings demonstrate the potential of MIR spectroscopy as a tool for assessment of AC integrity changes that result from injury.

Point-of-care diagnosis of tissue fibrosis: a review of advances in vibrational spectroscopy with machine learning

Histopathology is the gold standard for diagnosing fibrosis, but its routine use is constrained by the need for additional stains, time, personnel and resources. Vibrational spectroscopy is a novel technique that offers an alternative atraumatic approach, with short scan times, while providing metabolic and morphological data. This review evaluates vibrational spectroscopy for the assessment of fibrosis, with a focus on point-of-care capabilities. OVID Medline, Embase and Cochrane databases were systematically searched using PRISMA guidelines for search terms including vibrational spectroscopy, human tissue and fibrosis. Studies were stratified based on imaging modality and tissue type. Outcomes recorded included tissue type, machine learning technique, metrics for accuracy and author conclusions. Systematic review yielded 420 articles, of which 14 were relevant. Ten of these articles considered mid-infrared spectroscopy, three dealt with Raman spectroscopy and one with near-infrared spectroscopy. The metrics for detecting fibrosis were Pearson correlation coefficients ranging from 0.65-0.98; sensitivity from 76-100%; specificity from 90-99%; area under receiver operator curves from 0.83-0.98; and accuracy of 86-99%. Vibrational spectroscopy identified fibrosis in myeloproliferative neoplasms in bone, cirrhotic and hepatocellular carcinoma in liver, end-stage heart failure in cardiac tissue and following laser ablation for acne in skin. It also identified interstitial fibrosis as a predictor of early renal transplant rejection in renal tissue. Vibrational spectroscopic techniques can therefore accurately identify fibrosis in a range of human tissues. Emerging data show that it can be used to quantify, classify and provide data about the nature of fibrosis with a high degree of accuracy with potential scope for point-of-care use.

Towards a point-of-care multimodal spectroscopy instrument for the evaluation of human cardiac tissue

To demonstrate that point-of-care multimodal spectroscopy using Near-Infrared (NIR) and Raman Spectroscopy (RS) can be used to diagnose human heart tissue. We generated 105 spectroscopic scans, which comprised 4 NIR and 3 RS scans per sample to generate a "multimodal spectroscopic scan" (MSS) for each heart, done across 15 patients, 5 each from the dilated cardiomyopathy (DCM), Ischaemic Heart Disease (IHD) and Normal pathologies. Each of the MSS scans was undertaken in 3 s. Data were entered into machine learning (ML) algorithms to assess accuracy of MSS in diagnosing tissue type. The median age was 50 years (IQR 49-52) for IHD, 47 (IQR 45-50) for DCM and 36 (IQR 33-52) for healthy patients (p = 0.35), 60% of which were male. MSS identified key differences in IHD, DCM and normal heart samples in regions typically associated with fibrosis and collagen (NIR wavenumbers: 1433, 1509, 1581, 1689 and 1725 nm; RS wavelengths: 1658, 1450 and 1330 cm-1). In principal component (PC) analyses, these differences explained 99.2% of the variation in 4 PCs for NIR, 81.6% in 10 PCs for Raman, and 99.0% in 26 PCs for multimodal spectroscopic signatures. Using a stack machine learning algorithm with combined NIR and Raman data, our model had a precision of 96.9%, recall of 96.6%, specificity of 98.2% and Area Under Curve (AUC) of 0.989 (Table 1). NIR and Raman modalities alone had similar levels of precision at 94.4% and 89.8% respectively (Table 1). MSS combined with ML showed accuracy of 90% for detecting dilated cardiomyopathy, 100% for ischaemic heart disease and 100% for diagnosing healthy tissue. Multimodal spectroscopic signatures, based on NIR and Raman spectroscopy, could provide cardiac tissue scans in 3-s to aid accurate diagnoses of fibrosis in IHD, DCM and normal hearts. Table 1 Machine learning performance metrics for validation data sets of (a) Near-Infrared (NIR), (b) Raman and (c and d) multimodal data using logistic regression (LR), stochastic gradient descent (SGD) and support vector machines (SVM), with combined "stack" (LR + SGD + SVM) AUC Precision Recall Specificity (a) NIR model  Logistic regression 0.980 0.944 0.933 0.967  SGD 0.550 0.281 0.400 0.700  SVM 0.840 0.806 0.800 0.900  Stack 0.933 0.794 0.800 0.900 (b) Raman model  Logistic regression 0.985 0.940 0.929 0.960  SGD 0.892 0.869 0.857 0.932  SVM 0.992 0.940 0.929 0.960  Stack 0.954 0.869 0.857 0.932 (c) MSS: multimodal (NIR + Raman) to detect DCM vs. IHD vs. normal patients  Logistic regression 0.975 0.841 0.828 0.917  SGD 0.847 0.803 0.793 0.899  SVM 0.971 0.853 0.828 0.917  Stack 0.961 0.853 0.828 0.917 (d) MSS: multimodal (NIR + Raman) to detect pathological vs. normal patients  Logistic regression 0.961 0.969 0.966 0.984  SGD 0.944 0.967 0.966 0.923  SVM 1.000 1.000 1.000 1.000  Stack 1.000 0.944 0.931 0.969 Bold values indicate values obtained from the stack algorithm and used for analyses.

Point-of-care detection of fibrosis in liver transplant surgery using near-infrared spectroscopy and machine learning

Introduction

Visual assessment and imaging of the donor liver are inaccurate in predicting fibrosis and remain surrogates for histopathology. We demonstrate that 3-s scans using a handheld near-infrared-spectroscopy (NIRS) instrument can identify and quantify fibrosis in fresh human liver samples.

Methods

We undertook NIRS scans on 107 samples from 27 patients, 88 from 23 patients with liver disease, and 19 from four organ donors.

Results

Liver disease patients had a median immature fibrosis of 40% (interquartile range [IQR] 20-60) and mature fibrosis of 30% (10%-50%) on histopathology. The organ donor livers had a median fibrosis (both mature and immature) of 10% (IQR 5%-15%). Using machine learning, this study detected presence of cirrhosis and METAVIR grade of fibrosis with a classification accuracy of 96.3% and 97.2%, precision of 96.3% and 97.0%, recall of 96.3% and 97.2%, specificity of 95.4% and 98.0% and area under receiver operator curve of 0.977 and 0.999, respectively. Using partial-least square regression machine learning, this study predicted the percentage of both immature (R 2 = 0.842) and mature (R 2 = 0.837) with a low margin of error (root mean square of error of 9.76% and 7.96%, respectively).

Conclusion

This study demonstrates that a point-of-care NIRS instrument can accurately detect, quantify and classify liver fibrosis using machine learning.

Raman Spectroscopy and Machine Learning Enables Estimation of Articular Cartilage Structural, Compositional, and Functional Properties

OBJECTIVE

To differentiate healthy from artificially degraded articular cartilage and estimate its structural, compositional, and functional properties using Raman spectroscopy (RS).

DESIGN

Visually normal bovine patellae (n = 12) were used in this study. Osteochondral plugs (n = 60) were prepared and artificially degraded either enzymatically (via Collagenase D or Trypsin) or mechanically (via impact loading or surface abrasion) to induce mild to severe cartilage damage; additionally, control plugs were prepared (n = 12). Raman spectra were acquired from the samples before and after artificial degradation. Afterwards, reference biomechanical properties, proteoglycan (PG) content, collagen orientation, and zonal (%) thickness of the samples were measured. Machine learning models (classifiers and regressors) were then developed to discriminate healthy from degraded cartilage based on their Raman spectra and to predict the aforementioned reference properties.

RESULTS

The classifiers accurately categorized healthy and degraded samples (accuracy = 86%), and successfully discerned moderate from severely degraded samples (accuracy = 90%). On the other hand, the regression models estimated cartilage biomechanical properties with reasonable error (≤ 24%), with the lowest error observed in the prediction of instantaneous modulus (12%). With zonal properties, the lowest prediction errors were observed in the deep zone, i.e., PG content (14%), collagen orientation (29%), and zonal thickness (9%).

CONCLUSION

RS is capable of discriminating between healthy and damaged cartilage, and can estimate tissue properties with reasonable errors. These findings demonstrate the clinical potential of RS.

Visible and Near-Infrared Spectroscopy Enables Differentiation of Normal and Early Osteoarthritic Human Knee Joint Articular Cartilage

Osteoarthritis degenerates cartilage and impairs joint function. Early intervention opportunities are missed as current diagnostic methods are insensitive to early tissue degeneration. We investigated the capability of visible light-near-infrared spectroscopy (Vis-NIRS) to differentiate normal human cartilage from early osteoarthritic one. Vis-NIRS spectra, biomechanical properties and the state of osteoarthritis (OARSI grade) were quantified from osteochondral samples harvested from different anatomical sites of human cadaver knees. Two support vector machines (SVM) classifiers were developed based on the Vis-NIRS spectra and OARSI scores. The first classifier was designed to distinguish normal (OARSI: 0-1) from general osteoarthritic cartilage (OARSI: 2-5) to check the general suitability of the approach yielding an average accuracy of 75% (AUC = 0.77). Then, the second classifier was designed to distinguish normal from early osteoarthritic cartilage (OARSI: 2-3) yielding an average accuracy of 71% (AUC = 0.73). Important wavelength regions for differentiating normal from early osteoarthritic cartilage were related to collagen organization (wavelength region: 400-600 nm), collagen content (1000-1300 nm) and proteoglycan content (1600-1850 nm). The findings suggest that Vis-NIRS allows objective differentiation of normal and early osteoarthritic tissue, e.g., during arthroscopic repair surgeries.

Nondestructive Evaluation of Mechanical and Histological Properties of the Human Aorta With Near-Infrared Spectroscopy

INTRODUCTION

Ascending aortic dilatation is a well-known risk factor for aortic rupture. Indications for aortic replacement in its dilatation concomitant to other open-heart surgery exist; however, cut-off values based solely on aortic diameter may fail to identify patients with weakened aortic tissue. We introduce near-infrared spectroscopy (NIRS) as a diagnostic tool to nondestructively evaluate the structural and compositional properties of the human ascending aorta during open-heart surgeries. During open-heart surgery, NIRS could provide information regarding tissue viability in situ and thus contribute to the decision of optimal surgical repair.

MATERIALS AND METHODS

Samples were collected from patients with ascending aortic aneurysm (n = 23) undergoing elective aortic reconstruction surgery and from healthy subjects (n = 4). The samples were subjected to spectroscopic measurements, biomechanical testing, and histological analysis. The relationship between the near-infrared spectra and biomechanical and histological properties was investigated by adapting partial least squares regression.

RESULTS

Moderate prediction performance was achieved with biomechanical properties (r = 0.681, normalized root-mean-square error of cross-validation = 17.9%) and histological properties (r = 0.602, normalized root-mean-square error of cross-validation = 22.2%). Especially the performance with parameters describing the aorta's ultimate strength, for example, failure strain (r = 0.658), and elasticity (phase difference, r = 0.875) were promising and could, therefore, provide quantitative information on the rupture sensitivity of the aorta. For the estimation of histological properties, the results with α-smooth muscle actin (r = 0.581), elastin density (r = 0.973), mucoid extracellular matrix accumulation(r = 0.708), and media thickness (r = 0.866) were promising.

CONCLUSIONS

NIRS could be a potential technique for in situ evaluation of biomechanical and histological properties of human aorta and therefore useful in patient-specific treatment planning.

Dual-contrast micro-CT enables cartilage lesion detection and tissue condition evaluation ex vivo

BACKGROUND

Post-traumatic osteoarthritis is a frequent joint disease in the horse. Currently, equine medicine lacks effective methods to diagnose the severity of chondral defects after an injury.

OBJECTIVES

To investigate the capability of dual-contrast-enhanced computed tomography (dual-CECT) for detection of chondral lesions and evaluation of the severity of articular cartilage degeneration in the equine carpus ex vivo.

STUDY DESIGN

Pre-clinical experimental study.

METHODS

In nine Shetland ponies, blunt and sharp grooves were randomly created (in vivo) in the cartilage of radiocarpal and middle carpal joints. The contralateral joint served as control. The ponies were subjected to an 8-week exercise protocol and euthanised 39 weeks after surgery. CECT scanning (ex vivo) of the joints was performed using a micro-CT scanner 1 hour after an intra-articular injection of a dual-contrast agent. The dual-contrast agent consisted of ioxaglate (negatively charged, q = -1) and bismuth nanoparticles (BiNPs, q = 0, diameter ≈ 0.2 µm). CECT results were compared to histological cartilage proteoglycan content maps acquired using digital densitometry.

RESULTS

BiNPs enabled prolonged visual detection of both groove types as they are too large to diffuse into the cartilage. Furthermore, proportional ioxaglate diffusion inside the tissue allowed differentiation between the lesion and ungrooved articular cartilage (3 mm from the lesion and contralateral joint). The mean ioxaglate partition in the lesion was 19 percentage points higher (P < 0.001) when compared with the contralateral joint. The digital densitometry and the dual-contrast CECT findings showed good subjective visual agreement.

MAIN LIMITATIONS

Ex vivo study protocol and a low number of investigated joints.

CONCLUSIONS

The dual-CECT methodology, used in this study for the first time to image whole equine joints, is capable of effective lesion detection and simultaneous evaluation of the condition of the articular cartilage.

A Review of Machine Learning for Near-Infrared Spectroscopy

The analysis of infrared spectroscopy of substances is a non-invasive measurement technique that can be used in analytics. Although the main objective of this study is to provide a review of machine learning (ML) algorithms that have been reported for analyzing near-infrared (NIR) spectroscopy from traditional machine learning methods to deep network architectures, we also provide different NIR measurement modes, instruments, signal preprocessing methods, etc. Firstly, four different measurement modes available in NIR are reviewed, different types of NIR instruments are compared, and a summary of NIR data analysis methods is provided. Secondly, the public NIR spectroscopy datasets are briefly discussed, with links provided. Thirdly, the widely used data preprocessing and feature selection algorithms that have been reported for NIR spectroscopy are presented. Then, the majority of the traditional machine learning methods and deep network architectures that are commonly employed are covered. Finally, we conclude that developing the integration of a variety of machine learning algorithms in an efficient and lightweight manner is a significant future research direction.

Near-Infrared Spectroscopy Enables Arthroscopic Histologic Grading of Human Knee Articular Cartilage

Purpose

To develop the means to estimate cartilage histologic grades and proteoglycan content in ex vivo arthroscopy using near-infrared spectroscopy (NIRS).

Methods

In this experimental study, arthroscopic NIR spectral measurements were performed on both knees of 9 human cadavers, followed by osteochondral block extraction and in vitro measurements: reacquisition of spectra and reference measurements (proteoglycan content, and three histologic scores). A hybrid model, combining principal component analysis and linear mixed-effects model (PCA-LME), was trained for each reference to investigate its relationship with in vitro NIR spectra. The performance of the PCA-LME model was validated with ex vivo spectra before and after the exclusion of outlying spectra. Model performance was evaluated based on Spearman rank correlation (ρ) and root-mean-square error (RMSE).

Results

The PCA-LME models performed well (independent test: average ρ = 0.668, RMSE = 0.892, P < .001) in the prediction of the reference measurements based on in vitro data. The performance on ex vivo arthroscopic data was poorer but improved substantially after outlier exclusion (independent test: average ρ = 0.462 to 0.614, RMSE = 1.078 to 0.950, P = .019 to .008).

Conclusions

NIRS is capable of nondestructive evaluation of cartilage integrity (i.e., histologic scores and proteoglycan content) under similar conditions as in clinical arthroscopy.

Clinical Relevance

There are clear clinical benefits to the accurate assessment of cartilage lesions in arthroscopy. Visual grading is the current standard of care. However, optical techniques, such as NIRS, may provide a more objective assessment of cartilage damage.

Preclassification of Broadband and Sparse Infrared Data by Multiplicative Signal Correction Approach

Preclassification of raw infrared spectra has often been neglected in scientific literature. Separating spectra of low spectral quality, due to low signal-to-noise ratio, presence of artifacts, and low analyte presence, is crucial for accurate model development. Furthermore, it is very important for sparse data, where it becomes challenging to visually inspect spectra of different natures. Hence, a preclassification approach to separate infrared spectra for sparse data is needed. In this study, we propose a preclassification approach based on Multiplicative Signal Correction (MSC). The MSC approach was applied on human and the bovine knee cartilage broadband Fourier Transform Infrared (FTIR) spectra and on a sparse data subset comprising of only seven wavelengths. The goal of the preclassification was to separate spectra with analyte-rich signals (i.e., cartilage) from spectra with analyte-poor (and high-matrix) signals (i.e., water). The human datasets 1 and 2 contained 814 and 815 spectra, while the bovine dataset contained 396 spectra. A pure water spectrum was used as a reference spectrum in the MSC approach. A threshold for the root mean square error (RMSE) was used to separate cartilage from water spectra for broadband and the sparse spectral data. Additionally, standard noise-to-ratio and principle component analysis were applied on broadband spectra. The fully automated MSC preclassification approach, using water as reference spectrum, performed as well as the manual visual inspection. Moreover, it enabled not only separation of cartilage from water spectra in broadband spectral datasets, but also in sparse datasets where manual visual inspection cannot be applied.

Site- and Zone-Dependent Changes in Proteoglycan Content and Biomechanical Properties of Bluntly and Sharply Grooved Equine Articular Cartilage

In this study, we mapped and quantified changes of proteoglycan (PG) content and biomechanical properties in articular cartilage in which either blunt or sharp grooves had been made, both close to the groove and more remote of it, and at the opposing joint surface (kissing site) in equine carpal joints. In nine adult Shetland ponies, standardized blunt and sharp grooves were surgically made in the radiocarpal and middle carpal joints of a randomly chosen front limb. The contralateral control limb was sham-operated. At 39 weeks after surgery, ponies were euthanized. In 10 regions of interest (ROIs) (six remote from the grooves and four directly around the grooves), PG content as a function of tissue-depth and distance-to-groove was estimated using digital densitometry. Biomechanical properties of the cartilage were evaluated in the six ROIs remote from the grooves. Compared to control joints, whole tissue depth PG loss was found in sites adjacent to sharp and, to a larger extent, blunt grooves. Also, superficial PG loss of the surgically untouched kissing cartilage layers was observed. Significant PG loss was observed up to 300 µm (sharp) and at 500 µm (blunt) from the groove into the surrounding tissue. Equilibrium modulus was lower in grooved cartilage than in controls. Grooves, in particular blunt grooves, gave rise to severe PG loss close to the grooved sites and to mild degeneration more remote from the grooves in both sharply and bluntly grooved cartilage and at the kissing sites, resulting in loss of mechanical strength over the 9-month period.

Structural, compositional, and functional effects of blunt and sharp cartilage damage on the joint: A 9-month equine groove model study

This study aimed to quantify the long-term progression of blunt and sharp cartilage defects and their effect on joint homeostasis and function of the equine carpus. In nine adult Shetland ponies, the cartilage in the radiocarpal and middle carpal joint of one front limb was grooved (blunt or sharp randomized). The ponies were subjected to an 8-week exercise protocol and euthanized at 39 weeks. Structural and compositional alterations in joint tissues were evaluated in vivo using serial radiographs, synovial biopsies, and synovial fluid samples. Joint function was monitored by quantitative gait analysis. Macroscopic, microscopic, and biomechanical evaluation of the cartilage and assessment of subchondral bone parameters were performed ex vivo. Grooved cartilage showed higher OARSI microscopy scores than the contra-lateral sham-operated controls (p < 0.0001). Blunt-grooved cartilage scored higher than sharp-grooved cartilage (p = 0.007) and fixed charge density around these grooves was lower (p = 0.006). Equilibrium and instantaneous moduli trended lower in grooved cartilage than their controls (significant for radiocarpal joints). Changes in other tissues included a threefold to sevenfold change in interleukin-6 expression in synovium from grooved joints at week 23 (p = 0.042) and an increased CPII/C2C ratio in synovial fluid extracted from blunt-grooved joints at week 35 (p = 0.010). Gait analysis outcome revealed mild, gradually increasing lameness. In conclusion, blunt and, to a lesser extent, sharp grooves in combination with a period of moderate exercise, lead to mild degeneration in equine carpal cartilage over a 9-month period, but the effect on overall joint health remains limited.