From magnetic resonance spectroscopy to classification of tumors. A review of pattern recognition methods

In vivo brain MR spectroscopy in gliomas: clinical and pre-clinical chances

Purpose

Gliomas, the most common primary brain tumours, have recently been re-classified incorporating molecular aspects with important clinical, prognostic, and predictive implications. Concurrently, the reprogramming of metabolism, altering intracellular and extracellular metabolites affecting gene expression, differentiation, and the tumour microenvironment, is increasingly being studied, and alterations in metabolic pathways are becoming hallmarks of cancer. Magnetic resonance spectroscopy (MRS) is a complementary, non-invasive technique capable of quantifying multiple metabolites. The aim of this review focuses on the methodology and analysis techniques in proton MRS (1H MRS), including a brief look at X-nuclei MRS, and on its perspectives for diagnostic and prognostic biomarkers in gliomas in both clinical practice and preclinical research.

Methods

PubMed literature research was performed cross-linking the following key words: glioma, MRS, brain, in-vivo, human, animal model, clinical, pre-clinical, techniques, sequences, 1H, X-nuclei, Artificial Intelligence (AI), hyperpolarization.

Results

We selected clinical works (n = 51), preclinical studies (n = 35) and AI MRS application papers (n = 15) published within the last two decades. The methodological papers (n = 62) were taken into account since the technique first description.

Conclusions

Given the development of treatments targeting specific cancer metabolic pathways, MRS could play a key role in allowing non-invasive assessment for patient diagnosis and stratification, predicting and monitoring treatment responses and prognosis. The characterization of gliomas through MRS will benefit of a wide synergy among scientists and clinicians of different specialties within the context of new translational competences. Head coils, MRI hardware and post-processing analysis progress, advances in research, experts’ consensus recommendations and specific professionalizing programs will make the technique increasingly trustworthy, responsive, accessible.

The influence of sample collection, handling and low temperature storage upon NMR metabolic profiling analysis in human synovial fluid

The impact of metabolism upon the altered pathology of joint disease is rapidly becoming recognized as an important area of study. Synovial joint fluid is an attractive and representative biofluid of joint disease. A systemic review revealed little evidence of the metabolic stability of synovial joint fluid collection, handling or storage, despite recent reports characterizing the metabolic phenotype in joint disease. We aim to report the changes in small molecule detection within human synovial fluid (HSF) using nuclear magnetic resonance (NMR) spectroscopy at varying storage temperatures, durations and conditions. HSF was harvested by arthrocentesis from patients with isolated monoarthropathy or undergoing joint replacement (n = 30). Short-term storage (0-12 h, 4°C & 18°C) and the effect of repeated freeze-thaw cycles (-80°C to 18°C) was assessed. Long-term storage was evaluated by early (-80°C, <21days) and late analysis (-80°C, 10-12 months). 1D NMR spectroscopy experiments, NOESYGPPR1D and CPMG identified metabolites and semi-quantification was performed. Samples demonstrated broad stability to freeze-thaw cycling and refrigeration of <4 h. Short-term room temperature or refrigerated storage showed significant variation in 2-ketoisovalerate, valine, dimethylamine, succinate, 2-hydroxybutyrate, and acetaminophen glucuronide. Lipid and macromolecule detection was variable. Long-term storage demonstrated significant changes in: acetate, acetoacetate, creatine, N,N-dimethylglycine, dimethylsulfone, 3-hydroxybutyrate and succinate. Changeable metabolites during short-term storage appeared to be energy-synthesis intermediates. Most metabolites were stable for the first four hours at room temperature or refrigeration, with notable exceptions. We therefore recommend that HSF samples should be kept refrigerated for no more than 4 hours prior to freezing at -80°C. Furthermore, storage of HSF samples for 10-12 months before analysis can affect the detection of selected metabolites.

Can metabolic profiling provide a new description of osteoarthritis and enable a personalised medicine approach?

Osteoarthritis (OA) is a multifactorial disease contributing to significant disability and economic burden in Western populations. The aetiology of OA remains poorly understood, but is thought to involve genetic, mechanical and environmental factors. Currently, the diagnosis of OA relies predominantly on clinical assessment and plain radiographic changes long after the disease has been initiated. Recent advances suggest that there are changes in joint fluid metabolites that are associated with OA development. If this is the case, biochemical and metabolic biomarkers of OA could help determine prognosis, monitor disease progression and identify potential therapeutic targets. Moreover, for focussed management and personalised medicine, novel biomarkers could sub-stratify patients into OA phenotypes, differentiating metabolic OA from post-traumatic, age-related and genetic OA. To date, OA biomarkers have concentrated on cytokine action and protein signalling with some progress. However, these remain to be adopted into routine clinical practice. In this review, we outline the emerging metabolic links to OA pathogenesis and how an elucidation of the metabolic changes in this condition may provide future, more descriptive biomarkers to differentiate OA subtypes.

Determination of Grade and Receptor Status from the Primary Breast Lesion by Magnetic Resonance Spectroscopy

Magnetic resonance spectra (MRS) from fine needle aspiration biopsies (FNAB) from primary breast lesions were analysed using a pattern recognition method, Statistical Classification Strategy, to assess tumor grade and oestrogen receptor (ER) and progesterone receptor (PgR) status. Grade 1 and 2 breast cancers were separated from grade 3 cancers with a sensitivity and specificity of 96% and 95%, respectively. The ER status was predicted with a sensitivity of 91% and a specificity of 90%, and the PgR status with a sensitivity of 91% and a specificity of 86%. These classifiers provide rapid and reliable, computerized information and may offer an objective method for determining these prognostic indicators simultaneously with the diagnosis of primary pathology and lymph node involvement.

Metabolic mapping of deep brain structures and associations with symptomatology in autism spectrum disorders

Structural neuroimaging studies in autism report atypical volume in deep brain structures which are related to symptomatology. Little is known about metabolic changes in these regions, and how they vary with age and sex, and/or relate to clinical behaviors. Using magnetic resonance spectroscopy we measured N-acetylaspartate, choline, creatine, myoinositol and glutamate in the caudate, putamen, and thalamus of 20 children with autism and 16 typically developing controls (7-18 years). Relative to controls, individuals with autism had elevated glutamate/creatine in the putamen. In addition, both groups showed age-related increases in glutamate in this region. Boys, relative to girls had increased choline/creatine in the thalamus. Lastly, there were correlations between glutamate, choline, and myoinositol in all three regions, and behavioral scores in the ASD group. These findings suggest changes in deep gray matter neurochemistry, which are sensitive to diagnosis, age and sex, and are associated with behavioral differences.

Anatomic segmentation improves prostate cancer detection with artificial neural networks analysis of 1H magnetic resonance spectroscopic imaging

PURPOSE

To assess whether an artificial neural network (ANN) model is a useful tool for automatic detection of cancerous voxels in the prostate from (1)H-MRSI datasets and whether the addition of information about anatomical segmentation improves the detection of cancer.

MATERIALS AND METHODS

The Institutional Review Board approved this HIPAA-compliant study and waived informed consent. Eighteen men with prostate cancer (median age, 55 years; range, 36-71 years) who underwent endorectal MRI/MRSI before radical prostatectomy were included in this study. These patients had at least one cancer area on whole-mount histopathological map and at least one matching MRSI voxel suspicious for cancer detected. Two ANN models for automatic classification of MRSI voxels in the prostate were implemented and compared: model 1, which used only spectra as input, and model 2, which used the spectra plus information from anatomical segmentation. The models were trained, tested and validated using spectra from voxels that the spectroscopist had designated as cancer and that were verified on histopathological maps.

RESULTS

At ROC analysis, model 2 (AUC = 0.968) provided significantly better (P = 0.03) classification of cancerous voxels than did model 1 (AUC = 0.949).

CONCLUSION

Automatic analysis of prostate MRSI to detect cancer using ANN model is feasible. Application of anatomical segmentation from MRI as an additional input to ANN improves the accuracy of detecting cancerous voxels from MRSI.

Accurate classification of childhood brain tumours by in vivo ¹H MRS - a multi-centre study

AIMS

To evaluate the accuracy of single-voxel Magnetic Resonance Spectroscopy ((1)H MRS) as a non-invasive diagnostic aid for paediatric brain tumours in a multi-national study. Our hypotheses are (1) that automated classification based on (1)H MRS provides an accurate non-invasive diagnosis in multi-centre datasets and (2) using a protocol which increases the metabolite information improves the diagnostic accuracy.

METHODS

Seventy-eight patients under 16 years old with histologically proven brain tumours from 10 international centres were investigated. Discrimination of 29 medulloblastomas, 11 ependymomas and 38 pilocytic astrocytomas (PILOAs) was evaluated. Single-voxel MRS was undertaken prior to diagnosis (1.5 T Point-Resolved Spectroscopy (PRESS), Proton Brain Exam (PROBE) or Stimulated Echo Acquisition Mode (STEAM), echo time (TE) 20-32 ms and 135-136 ms). MRS data were processed using two strategies, determination of metabolite concentrations using TARQUIN software and automatic feature extraction with Peak Integration (PI). Linear Discriminant Analysis (LDA) was applied to this data to produce diagnostic classifiers. An evaluation of the diagnostic accuracy was performed based on resampling to measure the Balanced Accuracy Rate (BAR).

RESULTS

The accuracy of the diagnostic classifiers for discriminating the three tumour types was found to be high (BAR 0.98) when a combination of TE was used. The combination of both TEs significantly improved the classification performance (p<0.01, Tukey's test) compared with the use of one TE alone. Other tumour types were classified accurately as glial or primitive neuroectodermal (BAR 1.00).

CONCLUSION

(1)H MRS has excellent accuracy for the non-invasive diagnosis of common childhood brain tumours particularly if the metabolite information is maximised and should become part of routine clinical assessment for these children.

From cancer metabolism to new biomarkers and drug targets

Great interest is presently given to the analysis of metabolic changes that take place specifically in cancer cells. In this review we summarize the alterations in glycolysis, glutamine utilization, fatty acid synthesis and mitochondrial function that have been reported to occur in cancer cells and in human tumors. We then propose considering cancer as a system-level disease and argue how two hallmarks of cancer, enhanced cell proliferation and evasion from apoptosis, may be evaluated as system-level properties, and how this perspective is going to modify drug discovery. Given the relevance of the analysis of metabolism both for studies on the molecular basis of cancer cell phenotype and for clinical applications, the more relevant technologies for this purpose, from metabolome and metabolic flux analysis in cells by Nuclear Magnetic Resonance and Mass Spectrometry technologies to positron emission tomography on patients, are analyzed. The perspectives offered by specific changes in metabolism for a new drug discovery strategy for cancer are discussed and a survey of the industrial activity already going on in the field is reported.

Short-term temperature effect on the HRMAS spectra of human brain tumor biopsies and their pattern recognition analysis

OBJECT

To investigate the effect of temperature (0 versus 37 degrees C) in the high-resolution magic angle spinning spectroscopy (HRMAS) pattern of human brain tumor biopsies and its influence in recognition-based tumor type prediction. This proof-of-principle study addressed the bilateral discrimination between meningioma (MM) and glioblastoma multiforme (GBM) cases.

MATERIALS AND METHODS

Forty-three tumor biopsy samples were collected (20 MM and 23 GBM), kept frozen and later analyzed at 0 degrees C and 37 degrees C by HRMAS. Post-HRMAS histopathology was used to validate the tumor type. Time-course experiments (100 min) at both temperatures were carried out to monitor HRMAS pattern changes. Principal component analysis and linear discriminant analysis were used for classifier development with a training set of 20 biopsies.

RESULTS

Temperature-dependent, spectral pattern changes mostly affected mobile lipids and choline-containing compounds resonances and were essentially reversible. Incubation of 3 MM and 3 GBM at 37 degrees C during 100 minutes produced irreversible pattern changes below 13% in a few resonances. Classification performance of an independent test set of 7 biopsies was 100% for the pulse-and-acquire, CPMG at echo times (TE) of 30 ms and 144 ms and Hahn Echo at TE 30 ms at 0 degrees C and 37 degrees C. The performance for Hahn Echo spectra at 136 ms was 83.3% at 0 degrees C and 100% at 37 degrees C.

CONCLUSION

The spectral pattern of mobile lipids changes reversibly with temperature. HRMAS demonstrated potential for automated brain tumor biopsy classification. No advantage was obtained when acquiring spectra at 37 degrees C with respect to 0 degrees C in most of the conditions used for the discrimination addressed.

Molecular classification of brain tumor biopsies using solid-state magic angle spinning proton magnetic resonance spectroscopy and robust classifiers

Brain tumors are one of the leading causes of death in adults with cancer; however, molecular classification of these tumors with in vivo magnetic resonance spectroscopy (MRS) is limited because of the small number of metabolites detected. In vitro MRS provides highly informative biomarker profiles at higher fields, but also consumes the sample so that it is unavailable for subsequent analysis. In contrast, ex vivo high-resolution magic angle spinning (HRMAS) MRS conserves the sample but requires large samples and can pose technical challenges for producing accurate data, depending on the sample testing temperature. We developed a novel approach that combines a two-dimensional (2D), solid-state, HRMAS proton (1H) NMR method, TOBSY (total through-bond spectroscopy), which maximizes the advantages of HRMAS and a robust classification strategy. We used approximately 2 mg of tissue at -8 degrees C from each of 55 brain biopsies, and reliably detected 16 different biologically relevant molecular species. We compared two classification strategies, the support vector machine (SVM) classifier and a feed-forward neural network using the Levenberg-Marquardt back-propagation algorithm. We used the minimum redundancy/maximum relevance (MRMR) method as a powerful feature-selection scheme along with the SVM classifier. We suggest that molecular characterization of brain tumors based on highly informative 2D MRS should enable us to type and prognose even inoperable patients with high accuracy in vivo.

Multiproject-multicenter evaluation of automatic brain tumor classification by magnetic resonance spectroscopy

JUSTIFICATION

Automatic brain tumor classification by MRS has been under development for more than a decade. Nonetheless, to our knowledge, there are no published evaluations of predictive models with unseen cases that are subsequently acquired in different centers. The multicenter eTUMOUR project (2004-2009), which builds upon previous expertise from the INTERPRET project (2000-2002) has allowed such an evaluation to take place.

MATERIALS AND METHODS

A total of 253 pairwise classifiers for glioblastoma, meningioma, metastasis, and low-grade glial diagnosis were inferred based on 211 SV short TE INTERPRET MR spectra obtained at 1.5 T (PRESS or STEAM, 20-32 ms) and automatically pre-processed. Afterwards, the classifiers were tested with 97 spectra, which were subsequently compiled during eTUMOUR.

RESULTS

In our results based on subsequently acquired spectra, accuracies of around 90% were achieved for most of the pairwise discrimination problems. The exception was for the glioblastoma versus metastasis discrimination, which was below 78%. A more clear definition of metastases may be obtained by other approaches, such as MRSI + MRI.

CONCLUSIONS

The prediction of the tumor type of in-vivo MRS is possible using classifiers developed from previously acquired data, in different hospitals with different instrumentation under the same acquisition protocols. This methodology may find application for assisting in the diagnosis of new brain tumor cases and for the quality control of multicenter MRS databases.

The effect of combining two echo times in automatic brain tumor classification by MRS

(1)H MRS is becoming an accurate, non-invasive technique for initial examination of brain masses. We investigated if the combination of single-voxel (1)H MRS at 1.5 T at two different (TEs), short TE (PRESS or STEAM, 20-32 ms) and long TE (PRESS, 135-136 ms), improves the classification of brain tumors over using only one echo TE. A clinically validated dataset of 50 low-grade meningiomas, 105 aggressive tumors (glioblastoma and metastasis), and 30 low-grade glial tumors (astrocytomas grade II, oligodendrogliomas and oligoastrocytomas) was used to fit predictive models based on the combination of features from short-TEs and long-TE spectra. A new approach that combines the two consecutively was used to produce a single data vector from which relevant features of the two TE spectra could be extracted by means of three algorithms: stepwise, reliefF, and principal components analysis. Least squares support vector machines and linear discriminant analysis were applied to fit the pairwise and multiclass classifiers, respectively. Significant differences in performance were found when short-TE, long-TE or both spectra combined were used as input. In our dataset, to discriminate meningiomas, the combination of the two TE acquisitions produced optimal performance. To discriminate aggressive tumors from low-grade glial tumours, the use of short-TE acquisition alone was preferable. The classifier development strategy used here lends itself to automated learning and test performance processes, which may be of use for future web-based multicentric classifier development studies.

Spectrum separation resolves partial-volume effect of MRSI as demonstrated on brain tumor scans

Magnetic resonance spectroscopic imaging (MRSI) is currently used clinically in conjunction with anatomical MRI to assess the presence and extent of brain tumors and to evaluate treatment response. Unfortunately, the clinical utility of MRSI is limited by significant variability of in vivo spectra. Spectral profiles show increased variability because of partial coverage of large voxel volumes, infiltration of normal brain tissue by tumors, innate tumor heterogeneity, and measurement noise. We address these problems directly by quantifying the abundance (i.e. volume fraction) within a voxel for each tissue type instead of the conventional estimation of metabolite concentrations from spectral resonance peaks. This 'spectrum separation' method uses the non-negative matrix factorization algorithm, which simultaneously decomposes the observed spectra of multiple voxels into abundance distributions and constituent spectra. The accuracy of the estimated abundances is validated on phantom data. The presented results on 20 clinical cases of brain tumor show reduced cross-subject variability. This is reflected in improved discrimination between high-grade and low-grade gliomas, which demonstrates the physiological relevance of the extracted spectra. These results show that the proposed spectral analysis method can improve the effectiveness of MRSI as a diagnostic tool.

Mimicking the human expert: pattern recognition for an automated assessment of data quality in MR spectroscopic images

Besides the diagnostic evaluation of a spectrum, the assessment of its quality and a check for plausibility of its information remains a highly interactive and thus time-consuming process in MR spectroscopic imaging (MRSI) data analysis. In the automation of this quality control, a score is proposed that is obtained by training a machine learning classifier on a representative set of spectra that have previously been classified by experts into evaluable data and nonevaluable data. In the first quantitative evaluation of different quality measures on a test set of 45,312 long echo time spectra in the diagnosis of brain tumor, the proposed pattern recognition (using the random forest classifier) separated high- and low-quality spectra comparable to the human operator (area-under-the-curve of the receiver-operator-characteristic, AUC>0.993), and performed better than decision rules based on the signal-to-noise-ratio (AUC<0.934) or the estimated Cramér-Rao-bound on the errors of a spectral fitting (AUC<0.952). This probabilistic assessment of the data quality provides comprehensible confidence images and allows filtering the input of any subsequent data processing, i.e., quantitation or pattern recognition, in an automated fashion. It thus can increase robustness and reliability of the final diagnostic evaluation and allows for the automation of a tedious part of MRSI data analysis.

Technology insight: metabonomics in gastroenterology-basic principles and potential clinical applications

Metabonomics-the study of metabolic changes in an integrated biologic system-is an emerging field. This discipline joins the other 'omics' (genomics, transcriptomics and proteomics) to give rise to a comprehensive, systems-biology approach to the evaluation of holistic in vivo function. Metabonomics, especially when based on nuclear magnetic resonance spectroscopy, has the potential to identify biomarkers and prognostic factors, enhance clinical diagnosis, and expand hypothesis generation. As a consequence, the use of metabonomics has been extensively explored in the past decade, and applied successfully to the study of human diseases, toxicology, microbes, nutrition, and plant biology. This Review introduces the basic principles of nuclear magnetic resonance spectroscopy and commonly used tools for multivariate data analysis, before considering the applications and future potential of metabonomics in basic and clinical research, with emphasis on applications in the field of gastroenterology.

Correlation of magnetic resonance spectroscopic and growth characteristics within Grades II and III gliomas

OBJECT

The accurate diagnosis of World Health Organization Grades II and III gliomas is crucial for the effective treatment of patients with such lesions. Increased cell density and mitotic activity are histological features that distinguish Grade III from Grade II gliomas. Because increased cellular proliferation and density both contribute to the in vivo magnetic resonance (MR) spectroscopic peak corresponding to choline-containing compounds (Cho), the authors hypothesized that multivoxel MR spectroscopy might help identify the tumor regions with the most aggressive growth characteristics, which would be optimal locations for biopsy. They investigated the ability to use one or more MR spectroscopic parameters to predict the MIB-1 cell proliferation index (PI), the terminal deoxynucleotidyl transferase-mediated deoxyuridine triphosphate nick-end labeling cell death index (DI), the cell density, and the ratio of proliferation to cell death (PI/DI) within different regions of the same tumor.

METHODS

Patients with presumed Grades II or III glioma underwent 3D MR spectroscopic imaging prior to surgery, and two or three regions within the tumor were targeted for biopsy retrieval based on their spectroscopic features. Biopsy specimens were extracted from the tumor during image-guided resection, and the PI, DI, and cell density were assessed in the specimens using immunohistochemical methods.

CONCLUSIONS

The authors found that the relative levels of Cho and N-acetylaspartate (NAA) correlated with the cell density, PI, and PI/DI ratio within different regions of the same tumor and that the association held for the subpopulation of nonenhancing tumors. The association was stronger in tumors with large ranges of Cho/NAA values, irrespective of the presence of contrast enhancement. The findings demonstrate the validity of using MR spectroscopy to identify regions of aggressive growth in presumed Grade II or III gliomas that would be suitable targets for retrieving diagnostic biopsy specimens.

Automated estimation of tumor probability in prostate magnetic resonance spectroscopic imaging: pattern recognition vs quantification

Despite its diagnostic value and technological availability, (1)H NMR spectroscopic imaging (MRSI) has not found its way into clinical routine yet. Prerequisite for the clinical application is an automated and reliable method for the diagnostic evaluation of MRS images. In the present paper, different approaches to the estimation of tumor probability from MRSI in the prostate are assessed. Two approaches to feature extraction are compared: quantification (VARPRO, AMARES, QUEST) and subspace methods on spectral patterns (principal components, independent components, nonnegative matrix factorization, partial least squares). Linear as well as nonlinear classifiers (support vector machines, Gaussian processes, random forests) are applied and discussed. Quantification-based approaches are much more sensitive to the choice and parameterization of the quantification algorithm than to the choice of the classifier. Furthermore, linear methods based on magnitude spectra easily achieve equal performance and also allow for biochemical interpretation in combination with subspace methods. Nonlinear methods operating directly on magnitude spectra achieve the best results but are less transparent than the linear methods.

A metabolomics perspective of human brain tumours

During the past decade or so, a wealth of information about metabolites in various human brain tumour preparations (cultured cells, tissue specimens, tumours in vivo) has been accumulated by global profiling tools. Such holistic approaches to cellular biochemistry have been termed metabolomics. Inherent and specific metabolic profiles of major brain tumour cell types, as determined by proton nuclear magnetic resonance spectroscopy ((1)H MRS), have also been used to define metabolite phenotypes in tumours in vivo. This minireview examines the recent advances in the field of human brain tumour metabolomics research, including advances in MRS and mass spectrometry technologies, and data analysis.

Fast quantification of proton magnetic resonance spectroscopic imaging with artificial neural networks

Accurate quantification of the MRSI-observed regional distribution of metabolites involves relatively long processing times. This is particularly true in dealing with large amount of data that is typically acquired in multi-center clinical studies. To significantly shorten the processing time, an artificial neural network (ANN)-based approach was explored for quantifying the phase corrected (as opposed to magnitude) spectra. Specifically, in these studies radial basis function neural network (RBFNN) was used. This method was tested on simulated and normal human brain data acquired at 3T. The N-acetyl aspartate (NAA)/creatine (Cr), choline (Cho)/Cr, glutamate+glutamine (Glx)/Cr, and myo-inositol (mI)/Cr ratios in normal subjects were compared with the line fitting (LF) technique and jMRUI-AMARES analysis, and published values. The average NAA/Cr, Cho/Cr, Glx/Cr and mI/Cr ratios in normal controls were found to be 1.58+/-0.13, 0.9+/-0.08, 0.7+/-0.17 and 0.42+/-0.07, respectively. The corresponding ratios using the LF and jMRUI-AMARES methods were 1.6+/-0.11, 0.95+/-0.08, 0.78+/-0.18, 0.49+/-0.1 and 1.61+/-0.15, 0.78+/-0.07, 0.61+/-0.18, 0.42+/-0.13, respectively. These results agree with those published in literature. Bland-Altman analysis indicated an excellent agreement and minimal bias between the results obtained with RBFNN and other methods. The computational time for the current method was 15s compared to approximately 10 min for the LF-based analysis.

Optimal classification of long echo time in vivo magnetic resonance spectra in the detection of recurrent brain tumors

We describe the optimal high-level postprocessing of single-voxel (1)H magnetic resonance spectra and assess the benefits and limitations of automated methods as diagnostic aids in the detection of recurrent brain tumor. In a previous clinical study, 90 long-echo-time single-voxel spectra were obtained from 52 patients and classified during follow-up (30/28/32 normal/non-progressive tumor/tumor). Based on these data, a large number of evaluation strategies, including both standard resonance line quantification and algorithms from pattern recognition and machine learning, were compared in a quantitative evaluation. Results from linear and non-linear feature extraction, including ICA, PCA and wavelet transformations, and also the data from resonance line quantification were combined systematically with different classifiers such as LDA, chemometric methods (PLS, PCR), support vector machines and ensemble methods. Classification accuracy was assessed using a leave-one-out cross-validation scheme and the area under the curve (AUC) of the receiver operator characteristic (ROC). A regularized linear regression on spectra with binned channels reached 91% classification accuracy compared with 83% from quantification. Interpreting the loadings of these regressions, we find that lipid and lactate signals are too unreliable to be used in a simple machine rule. Choline and NAA are the main source of relevant information. Overall, we find that fully automated pattern recognition algorithms perform as well as, or slightly better than, a manually controlled and optimized resonance line quantification.

The Cinderella story of metabolic profiling: does metabolomics get to go to the functional genomics ball?

To date most global approaches to functional genomics have centred on genomics, transcriptomics and proteomics. However, since a number of high-profile publications, interest in metabolomics, the global profiling of metabolites in a cell, tissue or organism, has been rapidly increasing. A range of analytical techniques, including 1H NMR spectroscopy, gas chromatography-mass spectrometry (GC-MS), liquid chromatography-mass spectrometry (LC-MS), Fourier Transform mass spectrometry (FT-MS), high performance liquid chromatography (HPLC) and electrochemical array (EC-array), are required in order to maximize the number of metabolites that can be identified in a matrix. Applications have included phenotyping of yeast, mice and plants, understanding drug toxicity in pharmaceutical drug safety assessment, monitoring tumour treatment regimes and disease diagnosis in human populations. These successes are likely to be built on as other analytical and bioinformatic approaches are developed to fully exploit the information obtained in metabolic profiles. To assist in this process, databases of metabolomic data will be necessary to allow the passage of information between laboratories. In this prospective review, the capabilities of metabolomics in the field of medicine will be assessed in an attempt to predict the impact this 'Cinderella approach' will have at the 'functional genomic ball'.

Independent component analysis to proton spectroscopic imaging data of human brain tumours

In proton magnetic resonance spectroscopic imaging (1H MRSI), the recorded spectra are often linear combinations of spectra from different cell and tissue types within the voxel. This produces problems for data analysis and interpretation. A sophisticated approach is proposed here to handle the complexity of tissue heterogeneity in MRSI data. The independent component analysis (ICA) method was applied without prior knowledge to decompose the proton spectral components that relate to the heterogeneous cell populations with different proliferation and metabolism that are present in gliomas. The ability to classify brain tumours based on IC decomposite spectra was studied by grouping the components with histopathology. To this end, 10 controls and 34 patients with primary brain tumours were studied. The results indicate that ICA may reveal useful information from metabolic profiling for clinical purposes using long echo time MRSI of gliomas.

Homeostatic Signature of Anabolic Steroids in Cattle Using1H−13C HMBC NMR Metabonomics

We used metabonomics to discriminate the urinary signature of different anabolic steroid treatments in cattle having different physiological backgrounds (age, sex, and race). (1)H-(13)C heteronuclear multiple bonding connectivity NMR spectroscopy and multivariate statistical methods reveal that metabolites such as trimethylamine-N-oxide, dimethylamine, hippurate, creatine, creatinine, and citrate characterize the biological fingerprint of anabolic treatment. These urinary biomarkers suggest an overall homeostatic adaptation in nitrogen and energy metabolism. From results obtained in this study, it is now possible to consider metabonomics as a complementary method usable to improve doping control strategies to detect fraudulent anabolic treatment in cattle since the oriented global metabolic response provides helpful discrimination.

Melanoma Metastases in Regional Lymph Nodes Are Accurately Detected by Proton Magnetic Resonance Spectroscopy of Fine-Needle Aspirate Biopsy Samples

BACKGROUND

Nonsurgical assessment of sentinel nodes (SNs) would offer advantages over surgical SN excision by reducing morbidity and costs. Proton magnetic resonance spectroscopy (MRS) of fine-needle aspirate biopsy (FNAB) specimens identifies melanoma lymph node metastases. This study was undertaken to determine the accuracy of the MRS method and thereby establish a basis for the future development of a nonsurgical technique for assessing SNs.

METHODS

FNAB samples were obtained from 118 biopsy specimens from 77 patients during SN biopsy and regional lymphadenectomy. The specimens were histologically evaluated and correlated with MRS data. Histopathologic analysis established that 56 specimens contained metastatic melanoma and that 62 specimens were benign. A linear discriminant analysis-based classifier was developed for benign tissues and metastases.

RESULTS

The presence of metastatic melanoma in lymph nodes was predicted with a sensitivity of 92.9%, a specificity of 90.3%, and an accuracy of 91.5% in a primary data set. In a second data set that used FNAB samples separate from the original tissue samples, melanoma metastases were predicted with a sensitivity of 87.5%, a specificity of 90.3%, and an accuracy of 89.1%, thus supporting the reproducibility of the method.

CONCLUSIONS

Proton MRS of FNAB samples may provide a robust and accurate diagnosis of metastatic disease in the regional lymph nodes of melanoma patients. These data indicate the potential for SN staging of melanoma without surgical biopsy and histopathological evaluation.

Diagnosis of cerebral cryptococcoma using a computerized analysis of 1H NMR spectra in an animal model

Viable cryptococci load in biopsy material from an animal model of cerebral cryptococcoma were correlated with 1H NMR spectra and metabolite profiles. A statistical classification strategy was applied to distinguish among high-resolution 1H NMR spectra acquired from cryptococcomas, glioblastomas, and normal brain tissue. The overall classification accuracy was 100% when a genetic-algorithm-based optimal region selection preceded the development of linear discriminant analysis-based classifiers. The method remained robust despite differences in the microbial load of the cryptococcoma group when harvested at different time points. These results indicate the feasibility of the method for diagnosis without isolation of the pathogenic microorganism and its potential for in vivo diagnosis based on computerized analysis of magnetic resonance spectra.

Efficacy of proton magnetic resonance spectroscopy in neurological diagnosis and neurotherapeutic decision making

Anatomic and functional neuroimaging with magnetic resonance imaging (MRI) includes the technology more widely known as magnetic resonance spectroscopy (MRS). Now a routine automated "add-on" to all clinical magnetic resonance scanners, MRS, which assays regional neurochemical health and disease, is therefore the most accessible diagnostic tool for clinical management of neurometabolic disorders. Furthermore, the noninvasive nature of this technique makes it an ideal tool for therapeutic monitoring of disease and neurotherapeutic decision making. Among the more than 100 brain disorders that fall within this broad category, MRS contributes decisively to clinical decision making in a smaller but growing number. In this review, we will cover how MRS provides therapeutic impact in brain tumors, metabolic disorders such as adrenoleukodystrophy and Canavan's disease, Alzheimer's disease, hypoxia, secondary to trauma or ischemia, human immunodeficiency virus dementia and lesions, as well as systemic disease such as hepatic and renal failure. Together, these eight indications for MRS apply to a majority of all cases seen. This review, which examines the role of MRS in enhancing routine neurological practice and treatment concludes: 1) there is added value from MRS where MRI is positive; 2) there is unique decision-making information in MRS when MRI is negative; and 3) MRS usefully informs decision making in neurotherapeutics. Additional efficacy studies could extend the range of this capability.

Characterization of brain tumors by MRS, DWI and Ki-67 labeling index

With the advent of fast imaging hardware and specialized software, additional non-invasive magnetic resonance characterization of tumors has become available through proton magnetic resonance spectroscopy (MRS), hemodynamic imaging and diffusion-weighted imaging (DWI). Thus, patterns could be discerned to discriminate different types of tumors and even to infer their possible evolution in time. The purpose of this study was to investigate the correlation between MRS, DWI, histopathology and Ki-67 labeling index in a large number of brain tumors. Localized proton spectra were obtained in 47 patients with brain tumors who subsequently underwent surgery (biopsy or tumor removal). We performed MRS with short echo-time (30 ms) and metabolic values in spectra were measured using an external software with 25 peaks. In all patients who had DWI, we measured apparent diffusion coefficients (ADC) in the same region of interest (ROI) where the voxel in MRS was located. In most tumors the histological diagnosis and Ki-67 labeling index had been determined on our original surgical specimen. Cho/Cr, (Lip+Mm)/Cr, NAA/(Cho+Cr) and Glx/Cr indexes in MRS allowed discriminating between low- and high-grade gliomas and metastases (MTs). Likewise, absolute ADC values differentiated low- from high-grade gliomas expressed by Ki-67 labeling index. A novel finding was that high Glx/Cr in vivo MRS index (similar to other known indexes) was a good predictor of tumor grading.

Identification ofStaphylococcus aureusBrain Abscesses: Rat and Human Studies with1H MR Spectroscopy

PURPOSE

To determine the feasibility of a statistical classification strategy (SCS) and the identity of metabolites of bacterial and host origins that potentially contributed to the most discriminatory regions of magnetic resonance (MR) spectra from Staphylococcus aureus abscesses of biopsy material from controls, gliomas, and staphylococcal abscesses.

MATERIALS AND METHODS

Human and animal study received ethics committee approval, and informed patient consent was obtained. A rat model of S aureus brain abscess was developed. Histologic and microbiologic examination was performed to assess abscess development 3-4, 6-8, and 10-15 days after initiation. Metabolite profiles in pus (n = 62) and controls (n = 37) were characterized with ex vivo MR spectroscopy and compared with data from rat gliomas (n = 27). SCS, optimal region selection, and development of pairwise classifiers allowed MR spectra of abscesses (n = 42, day 6-8) to be distinguished from those of glioblastoma multiforme and controls. MR spectroscopy profiles of pus from animal abscesses were compared with in vivo MR spectra from patients with staphylococcal brain abscesses (n = 7, aged 6-67 years) and ex vivo pus MR spectra from patients with S aureus abscesses.

RESULTS

Histologically confirmed abscesses were present 6-8 days after stereotactic injection of S aureus in 42 of 47 rats (89%). MR spectra of abscesses and glioblastoma multiforme in the animal model were similar. Typical metabolites of abscesses due to anaerobe bacteria (acetate, succinate, amino acids) were not detectable in S aureus abscesses in rats or humans. MR spectroscopic findings from controls, abscesses, and gliomas were distinguished by means of SCS with an accuracy of 99%. Analysis of the most discriminatory regions with two-dimensional correlation spectra indicated that glutamine and/or glutamate and aspartate potentially contributed to successful classification.

CONCLUSION

S aureus is detectable in abscesses with a non-culture-based method in an animal model.

Combination of feature-reduced MR spectroscopic and MR imaging data for improved brain tumor classification

The purpose of this paper is to evaluate the effect of the combination of magnetic resonance spectroscopic imaging (MRSI) data and magnetic resonance imaging (MRI) data on the classification result of four brain tumor classes. Suppressed and unsuppressed short echo time MRSI and MRI were performed on 24 patients with a brain tumor and four volunteers. Four different feature reduction procedures were applied to the MRSI data: simple quantitation, principal component analysis, independent component analysis and LCModel. Water intensities were calculated from the unsuppressed MRSI data. Features were extracted from the MR images which were acquired with four different contrasts to comply with the spatial resolution of the MRSI. Evaluation was performed by investigating different combinations of the MRSI features, the MRI features and the water intensities. For each data set, the isolation in feature space of the tumor classes, healthy brain tissue and cerebrospinal fluid was calculated and visualized. A test set was used to calculate classification results for each data set. Finally, the effect of the selected feature reduction procedures on the MRSI data was investigated to ascertain whether it was more important than the addition of MRI information. Conclusions are that the combination of features from MRSI data and MRI data improves the classification result considerably when compared with features obtained from MRSI data alone. This effect is larger than the effect of specific feature reduction procedures on the MRSI data. The addition of water intensities to the data set also increases the classification result, although not significantly. We show that the combination of data from different MR investigations can be very important for brain tumor classification, particularly if a large number of tumors are to be classified simultaneously.

Automatic alignment of individual peaks in large high-resolution spectral data sets

Pattern recognition techniques are effective tools for reducing the information contained in large spectral data sets to a much smaller number of significant features which can then be used to make interpretations about the chemical or biochemical system under study. Often the effectiveness of such approaches is impeded by experimental and instrument induced variations in the position, phase, and line width of the spectral peaks. Although characterizing the cause and magnitude of these fluctuations could be important in its own right (pH-induced NMR chemical shift changes, for example) in general they obscure the process of pattern discovery. One major area of application is the use of large databases of (1)H NMR spectra of biofluids such as urine for investigating perturbations in metabolic profiles caused by drugs or disease, a process now termed metabonomics. Frequency shifts of individual peaks are the dominant source of such unwanted variations in this type of data. In this paper, an automatic procedure for aligning the individual peaks in the data set is described and evaluated. The proposed method will be vital for the efficient and automatic analysis of large metabonomic data sets and should also be applicable to other types of data.

Classification of neck movement patterns related to whiplash-associated disorders using neural networks

This paper presents a new method for classification of neck movement patterns related to Whiplash-associated disorders (WAD) using a resilient backpropagation neural network (BPNN). WAD are a common diagnosis after neck trauma, typically caused by rear-end car accidents. Since physical injuries seldom are found with present imaging techniques, the diagnosis can be difficult to make. The active range of the neck is often visually inspected in patients with neck pain, but this is a subjective measure, and a more objective decision support system, that gives a reliable and more detailed analysis of neck movement pattern, is needed. The objective of this study was to evaluate the predictive ability of a BPNN, using neck movement variables as input. Three-dimensional (3-D) neck movement data from 59 subjects with WAD and 56 control subjects were collected with a ProReflex system. Rotation angle and angle velocity were calculated using the instantaneous helical axis method and motion variables were extracted. A principal component analysis was performed in order to reduce data and improve the BPNN performance. BPNNs with six hidden nodes had a predictivity of 0.89, a sensitivity of 0.90 and a specificity of 0.88, which are very promising results. This shows that neck movement analysis combined with a neural network could build the basis of a decision support system for classifying suspected WAD, even though further evaluation of the method is needed.

A Chemometric Approach for Brain Tumor Classification Using Magnetic Resonance Imaging and Spectroscopy

A new classification approach was developed to improve the noninvasive diagnosis of brain tumors. Within this approach, information is extracted from magnetic resonance imaging and spectroscopy data, from which the relative location and distribution of selected tumor classes in feature space can be calculated. This relative location and distribution is used to select the best information extraction procedure, to identify overlapping tumor classes, and to calculate probabilities of class membership. These probabilities are very important, since they provide information about the reliability of classification and might provide information about the heterogeneity of the tissue. Classification boundaries were calculated by setting thresholds for each investigated tumor class, which enabled the classification of new objects. Results on histopathologically determined tumors are excellent, demonstrated by spatial maps showing a high probability for the correctly identified tumor class and, moreover, low probabilities for other tumor classes.

Rapid identification of Candida species by using nuclear magnetic resonance spectroscopy and a statistical classification strategy

Nuclear magnetic resonance (NMR) spectra were acquired from suspensions of clinically important yeast species of the genus Candida to characterize the relationship between metabolite profiles and species identification. Major metabolites were identified by using two-dimensional correlation NMR spectroscopy. One-dimensional proton NMR spectra were analyzed by using a staged statistical classification strategy. Analysis of NMR spectra from 442 isolates of Candida albicans, C. glabrata, C. krusei, C. parapsilosis, and C. tropicalis resulted in rapid, accurate identification when compared with conventional and DNA-based identification. Spectral regions used for the classification of the five yeast species revealed species-specific differences in relative amounts of lipids, trehalose, polyols, and other metabolites. Isolates of C. parapsilosis and C. glabrata with unusual PCR fingerprinting patterns also generated atypical NMR spectra, suggesting the possibility of intraspecies discontinuity. We conclude that NMR spectroscopy combined with a statistical classification strategy is a rapid, nondestructive, and potentially valuable method for identification and chemotaxonomic characterization that may be broadly applicable to fungi and other microorganisms.

Diagnostic potential of near-infrared Raman spectroscopy in the colon: differentiating adenomatous from hyperplastic polyps

BACKGROUND

Near-infrared Raman spectroscopy is a promising optical technique for GI tissue diagnosis. This study assessed the diagnostic potential of near-infrared Raman spectroscopy in the colon by evaluating its ability to distinguish between adenomatous and hyperplastic polyps.

METHODS

Ex vivo and in vivo Raman spectra of colon polyps were collected by using a custom-built, fiber-optic, near-infrared Raman spectroscopic system. Multivariate statistical techniques, including principal component analysis and linear discriminant analysis, were used to develop diagnostic algorithms for classifying colon polyps based on their spectral characteristics. With the number of samples available, spectral classification of polyps was tested by using a leave-one-out, cross-validation method.

RESULTS

Fifty-four ex vivo Raman spectra were analyzed (20 hyperplastic, 34 adenomatous). The spectral-based diagnostic algorithms identified adenomatous polyps with 91% sensitivity, 95% specificity, and 93% accuracy. In vivo, adenomas (n = 10) were distinguished from hyperplastic polyps (n = 9) with 100% sensitivity, 89% specificity, and 95% accuracy.

CONCLUSIONS

Near-infrared Raman spectroscopy differentiated adenomatous from hyperplastic polyps with high diagnostic accuracy. To our knowledge, this is the first demonstration of the potential of near-infrared Raman spectroscopy for differentiation of colonic polyps during GI endoscopy.

Automated classification of short echo time in in vivo 1H brain tumor spectra: a multicenter study

Automated pattern recognition techniques are needed to help radiologists categorize MRS data of brain tumors according to histological type and grade. A major question is whether a computer program "trained" on spectra from one hospital will be able to classify those from another, particularly if the acquisition protocol is different. A subset of 144 histopathologically validated brain tumor spectra in the INTERPRET database, obtained from three of the collaborating centers, was grouped into meningiomas, low-grade astrocytomas, and "aggressive tumors" (glioblastomas and metastases). Spectra from two centers formed the training set (94 spectra) while the third acted as the test set (50 spectra). Linear discriminant analysis successfully classified 48/50 in the test set; the remaining two were atypical cases. When the training and test sets were combined, 133 of the 144 spectra were correctly classified using the leave-one-out procedure. These spectra had been obtained using different sequences (STEAM and PRESS), different echo times (20, 30, 31, and 32 ms), different repetition times (1600 and 2000 ms), and different manufacturers' instruments (GE and Philips). Pattern recognition algorithms are less sensitive to acquisition parameters than had been expected.

Pathology of hepatocellular carcinoma and its precursors using proton magnetic resonance spectroscopy and a statistical classification strategy

AIM

Apply the statistical classification strategy (SCS) to magnetic resonance spectroscopy (MRS) data from liver biopsies and test its potential to discriminate between normal liver, cirrhotic nodules and nodules of hepatocellular carcinoma with a high degree of accuracy.

METHODS

Liver tissue specimens from 54 patients undergoing either partial (hemi) or total hepatectomy were analysed by one-dimensional proton MRS at 8.5 Tesla. Histologically, these specimens were confirmed as normal (n=31), cirrhotic (n=59), and hepatocellular carcinoma (HCC, n=32). Diagnostic correlation was performed between the MR spectra and histopathology. An SCS was applied consisting of pre-processing MR magnitude spectra to identify spectral regions of maximal discriminatory value, and cross-validated linear discriminant analysis.

RESULTS

SCS applied to MRS data distinguished normal liver tissue from HCC with an accuracy of 100%. Normal liver tissue was distinguished from cirrhotic liver with an accuracy of 92% and cirrhotic liver was distinguished from HCC with an accuracy of 98%.

CONCLUSIONS

SCS applied to proton MRS of liver biopsies provides a robust method to distinguish, with a high degree of accuracy, HCC from both cirrhotic and normal liver.

Applications of neural network analyses to in vivo 1H magnetic resonance spectroscopy of Parkinson disease patients

PURPOSE

To apply neural network analyses to in vivo magnetic resonance spectra of controls and Parkinson disease (PD) patients for the purpose of classification.

MATERIALS AND METHODS

Ninety-seven in vivo proton magnetic resonance spectra of the basal ganglia were recorded from 31 patients with (PD) and 14 age-matched healthy volunteers on a 1.5-T imager. The PD patients were grouped as follows: probable PD (N = 15), possible PD (N = 11), and atypical PD (N = 5). Total acquisition times of approximately five minutes were achieved with a TE (echo time) of 135 msec, a TR (repetition time) of 2000 msec, and 128 scan averages. Neural network (back propagation, Kohonen, probabilistic, and radial basis function) and related (generative topographic mapping) data analyses were performed.

RESULTS

Conventional data analysis showed no statistically significant differences in metabolite ratios based on measuring signal intensities. The trained networks could distinguish control from PD with considerable accuracy (true positive fraction 0.971, true negative fraction 0.933). When four classes were defined, approximately 88% of the predictions were correct. The multivariate analysis indicated metabolic changes in the basal ganglia in PD.

CONCLUSION

A variety of neural network and related approaches can be successfully applied to both qualitative visualization and classification of in vivo spectra of PD patients.

Metabonomic assessment of physiological disruptions using 1H-13C HMBC-NMR spectroscopy combined with pattern recognition procedures performed on filtered variables

Metabonomic characterization of long-lasting although weak physiological events such as anabolic disruptions remains poorly investigated. We have validated 1H-13C HMBC-NMR as a suitable generator of instrumental variables that are strongly linked to the concentration of endogenous metabolites in biological fluids. This method is interfaced to multivariate pattern recognition procedures. Fingerprints established from urine sample collected on cattle treated with anabolic steroids were used to validate this method. Four main results arise from this study. (i) 2D NMR is as informative as 1D NMR. (ii) 2D NMR variable clustering highlights successfully a contingent redundancy of variables, although a relevant hierarchical model of statistical correlations covering from structural relationships to physiologic links can also be evidenced. (iii) To enhance pattern recognition performances, we have validated a variable selection algorithm for accurate prediction of unknown individuals belonging to predetermined groups achieved by linear discriminant analysis (LDA). This algorithm synthesizes the whole information contained in the data set by selecting preferentially nonredundant variables. Parameters generating variable subsets are validated by predicted variance efficiency obtained when minimizing error rates calculated by cross-validation methods. (iv) Provided variables are correctly filtered, LDA fairly competes with partial least-squares methods for both classification of individuals and statistical interpretation of metabolic responses obtained in such a physiological disruption context.

Interpretation of static time-of-flight secondary ion mass spectra of adsorbed protein films by multivariate pattern recognition

Multivariate analysis has become increasingly common in the analysis of multidimensional spectral data. We previously showed that the multivariate analysis technique principal component analysis (PCA) is an excellent method for interpreting the static time-of-flight secondary ion mass spectrometry (TOF-SIMS) spectra of adsorbed protein films. PCA is an unsupervised pattern recognition technique that loses resolution between spectra of different proteins as more proteins are added to the data set due to large within-group variation. The supervised pattern recognition techniques discriminant principal component analysis (DPCA) and linear discriminant analysis (LDA), which aim to control within-group variation while maximizing between-group separation to enhance discrimination between groups, were compared with PCA using data sets of TOF-SIMS spectra of proteins adsorbed onto mica and PTFE substrates. DPCA and LDA quantitatively improved discrimination between groups and provided different information about the data than PCA. LDA was able to classify unknown samples with a misclassification rate lower than PCA or DPCA. Both unsupervised and supervised pattern recognition techniques are useful for the interpretation and classification of static TOF-SIMS spectra of adsorbed protein films.

Novel approaches to imaging brain tumors

Brain imaging techniques are assuming a greater range of roles in neuro-oncology. New techniques promise earlier recognition of the spread of tumors to the brain, which is useful in staging of disseminated disease, as well as better definition of small lesions associated with presentations of epilepsy. There is the promise that entirely noninvasive, specific diagnosis of brain tumors may become possible. Imaging methods are being used increasingly to direct and monitor therapy. Preoperative and intraoperative imaging are being used for guiding tumor surgery. An exciting potential goal for greater use of imaging is in the individualization of medical therapies either by analysis of in vitro responses or by visualization of drug responses on the tumor in situ. An important focus for technical development is in the robust integration of complementary information to allow optimization of the sensitivity and specificity of multimodal examinations.