Distinguishing between paediatric brain tumour types using multi-parametric magnetic resonance imaging and machine learning: A multi-site study.
ABSTRACT: The imaging and subsequent accurate diagnosis of paediatric brain tumours presents a radiological challenge, with magnetic resonance imaging playing a key role in providing tumour specific imaging information. Diffusion weighted and perfusion imaging are commonly used to aid the non-invasive diagnosis of children's brain tumours, but are usually evaluated by expert qualitative review. Quantitative studies are mainly single centre and single modality. The aim of this work was to combine multi-centre diffusion and perfusion imaging, with machine learning, to develop machine learning based classifiers to discriminate between three common paediatric tumour types. The results show that diffusion and perfusion weighted imaging of both the tumour and whole brain provide significant features which differ between tumour types, and that combining these features gives the optimal machine learning classifier with >80% predictive precision. This work represents a step forward to aid in the non-invasive diagnosis of paediatric brain tumours, using advanced clinical imaging.
Project description:BACKGROUND:Diffusion- and perfusion-weighted MRI are valuable tools for measuring the cellular and vascular properties of brain tumours. This has been well studied in adult patients, however, the biological features of childhood brain tumours are unique, and paediatric-focused studies are less common. We aimed to assess the diagnostic utility of apparent diffusion coefficient (ADC) values derived from diffusion-weighted imaging (DWI) and cerebral blood flow (CBF) values derived from arterial spin labelling (ASL) in paediatric brain tumours. METHODS:We performed a meta-analysis of published studies reporting ADC and ASL-derived CBF values in paediatric brain tumours. Data were combined using a random effects model in order to define typical parameter ranges for different histological tumour subtypes and WHO grades. New data were also acquired in a 'validation cohort' at our institution, in which ADC and CBF values in treatment naïve paediatric brain tumour patients were measured, in order to test the validity of the findings from the literature in an un-seen cohort. ADC and CBF quantification was performed by two radiologists via manual placement of tumour regions of interest (ROIs), in addition to an automated approach to tumour ROI placement. RESULTS:A total of 14 studies met the inclusion criteria for the meta-analysis, constituting data acquired in 542 paediatric patients. Parameters of interest were based on measurements from ROIs placed within the tumour, including mean and minimum ADC values (ADCROI-mean, ADCROI-min) and the maximum CBF value normalised to grey matter (nCBFROI-max). After combination of the literature data, a number of histological tumour subtype groups showed significant differences in ADC values, which were confirmed, where possible, in our validation cohort of 32 patients. In both the meta-analysis and our cohort, diffuse midline glioma was found to be an outlier among high-grade tumour subtypes, with ADC and CBF values more similar to the low-grade tumours. After grouping patients by WHO grade, significant differences in grade groups were found in ADCROI-mean, ADCROI-min, and nCBFROI-max, in both the meta-analysis and our validation cohort. After excluding diffuse midline glioma, optimum thresholds (derived from ROC analysis) for separating low/high-grade tumours were 0.95 × 10-3 mm2/s (ADCROI-mean), 0.82 × 10-3 mm2/s (ADCROI-min) and 1.45 (nCBFROI-max). These thresholds were able to identify low/high-grade tumours with 96%, 83%, and 83% accuracy respectively in our validation cohort, and agreed well with the results from the meta-analysis. Diagnostic power was improved by combining ADC and CBF measurements from the same tumour, after which 100% of tumours in our cohort were correctly classified as either low- or high-grade (excluding diffuse midline glioma). CONCLUSION:ADC and CBF values are useful for differentiating certain histological subtypes, and separating low- and high-grade paediatric brain tumours. The threshold values presented here are in agreement with previously published studies, as well as a new patient cohort. If ADC and CBF values acquired in the same tumour are combined, the diagnostic accuracy is optimised.
Project description:Routine blood test results are assumed to contain much more information than is usually recognised even by the most experienced clinicians. Using routine blood tests from 15,176 neurological patients we built a machine learning predictive model for the diagnosis of brain tumours. We validated the model by retrospective analysis of 68 consecutive brain tumour and 215 control patients presenting to the neurological emergency service. Only patients with head imaging and routine blood test data were included in the validation sample. The sensitivity and specificity of the adapted tumour model in the validation group were 96% and 74%, respectively. Our data demonstrate the feasibility of brain tumour diagnosis from routine blood tests using machine learning. The reported diagnostic accuracy is comparable and possibly complementary to that of imaging studies. The presented machine learning approach opens a completely new avenue in the diagnosis of these grave neurological diseases and demonstrates the utility of valuable information obtained from routine blood tests.
Project description:PRIMAGE is one of the largest and more ambitious research projects dealing with medical imaging, artificial intelligence and cancer treatment in children. It is a 4-year European Commission-financed project that has 16 European partners in the consortium, including the European Society for Paediatric Oncology, two imaging biobanks, and three prominent European paediatric oncology units. The project is constructed as an observational in silico study involving high-quality anonymised datasets (imaging, clinical, molecular, and genetics) for the training and validation of machine learning and multiscale algorithms. The open cloud-based platform will offer precise clinical assistance for phenotyping (diagnosis), treatment allocation (prediction), and patient endpoints (prognosis), based on the use of imaging biomarkers, tumour growth simulation, advanced visualisation of confidence scores, and machine-learning approaches. The decision support prototype will be constructed and validated on two paediatric cancers: neuroblastoma and diffuse intrinsic pontine glioma. External validation will be performed on data recruited from independent collaborative centres. Final results will be available for the scientific community at the end of the project, and ready for translation to other malignant solid tumours.
Project description:BACKGROUND:Improved outcome prediction is vital for the delivery of risk-adjusted, appropriate and effective care to paediatric patients with Ewing sarcoma-the second most common paediatric malignant bone tumour. Fourier transform infrared (FTIR) spectroscopy of tissues allows the bulk biochemical content of a biological sample to be probed and makes possible the study and diagnosis of disease. METHODS:In this retrospective study, FTIR spectra of sections of biopsy-obtained bone tissue were recorded. Twenty-seven patients (between 5 and 20 years of age) with newly diagnosed Ewing sarcoma of bone were included in this study. The prognostic value of FTIR spectra obtained from Ewing sarcoma (ES) tumours before and after neoadjuvant chemotherapy were analysed in combination with various data-reduction and machine learning approaches. RESULTS:Random forest and linear discriminant analysis supervised learning models were able to correctly predict patient mortality in 92% of cases using leave-one-out cross-validation. The best performing model for predicting patient relapse was a linear Support Vector Machine trained on the observed spectral changes as a result of chemotherapy treatment, which achieved 92% accuracy. CONCLUSION:FTIR spectra of tumour biopsy samples may predict treatment outcome in paediatric Ewing sarcoma patients with greater than 92% accuracy.
Project description:Currently, breast cancer often requires invasive biopsies for diagnosis, motivating researchers to design and develop non-invasive and automated diagnosis systems. Recent microwave breast imaging studies have shown how backscattered signals carry relevant information about the shape of a tumour, and tumour shape is often used with current imaging modalities to assess malignancy. This paper presents a comprehensive analysis of microwave breast diagnosis systems which use machine learning to learn characteristics of benign and malignant tumours. The state-of-the-art, the main challenges still to overcome and potential solutions are outlined. Specifically, this work investigates the benefit of signal pre-processing on diagnostic performance, and proposes a new set of extracted features that capture the tumour shape information embedded in a signal. This work also investigates if a relationship exists between the antenna topology in a microwave system and diagnostic performance. Finally, a careful machine learning validation methodology is implemented to guarantee the robustness of the results and the accuracy of performance evaluation.
Project description:INTRODUCTION:In recent years, numerous methods have been proposed to predict tissue outcome in acute stroke patients using machine learning methods incorporating multiparametric imaging data. Most methods include diffusion and perfusion parameters as image-based parameters but do not include any spatial information although these parameters are spatially dependent, e.g. different perfusion properties in white and gray brain matter. This study aims to investigate if including spatial features improves the accuracy of multi-parametric tissue outcome prediction. MATERIALS AND METHODS:Acute and follow-up multi-center MRI datasets of 99 patients were available for this study. Logistic regression, random forest, and XGBoost machine learning models were trained and tested using acute MR diffusion and perfusion features and known follow-up lesions. Different combinations of atlas coordinates and lesion probability maps were included as spatial information. The stroke lesion predictions were compared to the true tissue outcomes using the area under the receiver operating characteristic curve (ROC AUC) and the Dice metric. RESULTS:The statistical analysis revealed that including spatial features significantly improves the tissue outcome prediction. Overall, the XGBoost and random forest models performed best in every setting and achieved state-of-the-art results regarding both metrics with similar improvements achieved including Montreal Neurological Institute (MNI) reference space coordinates or voxel-wise lesion probabilities. CONCLUSION:Spatial features should be integrated to improve lesion outcome prediction using machine learning models.
Project description:The goal of assessing tumour response on imaging is to identify patients who are likely to benefit - or not - from anticancer treatment, especially in relation to survival. The World Health Organization was the first to develop assessment criteria. This early score, which assessed tumour burden by standardising lesion size measurements, laid the groundwork for many of the criteria that followed. This was then improved by the Response Evaluation Criteria in Solid Tumours (RECIST) which was quickly adopted by the oncology community. At the same time, many interventional oncology treatments were developed to target specific features of liver tumours that result in significant changes in tumours but have little effect on tumour size. New criteria focusing on the viable part of tumours were therefore designed to provide more appropriate feedback to guide patient management. Targeted therapy has resulted in a breakthrough that challenges conventional response criteria due to the non-linear relationship between response and tumour size, requiring the development of methods that emphasize the appearance of tumours. More recently, research into functional and quantitative imaging has created new opportunities in liver imaging. These results have suggested that certain parameters could serve as early predictors of response or could predict later tumour response at baseline. These approaches have now been extended by machine learning and deep learning. This clinical review focuses on the progress made in the evaluation of liver tumours on imaging, discussing the rationale for this approach, addressing challenges and controversies in the field, and suggesting possible future developments.
Project description:<h4>Objectives</h4>Hypoxia is associated with poor prognosis and treatment resistance in breast cancer. However, the temporally variant nature of hypoxia can complicate interpretation of imaging findings. We explored the relationship between hypoxia and vascular function in breast tumours through combined <sup>18</sup>F-fluoromisonidazole (<sup>18</sup> F-FMISO) PET/MRI, with simultaneous assessment circumventing the effect of temporal variation in hypoxia and perfusion.<h4>Methods</h4>Women with histologically confirmed, primary breast cancer underwent a simultaneous <sup>18</sup>F-FMISO-PET/MR examination. Tumour hypoxia was assessed using influx rate constant K<sub>i</sub> and hypoxic fractions (%HF), while parameters of vascular function (K<sup>trans</sup>, k<sub>ep</sub>, v<sub>e</sub>, v<sub>p</sub>) and cellularity (ADC) were derived from dynamic contrast-enhanced (DCE) and diffusion-weighted (DW)-MRI, respectively. Additional correlates included histological subtype, grade and size. Relationships between imaging variables were assessed using Pearson correlation (r).<h4>Results</h4>Twenty-nine women with 32 lesions were assessed. Hypoxic fractions >?1% were observed in 6/32 (19%) cancers, while 18/32 (56%) tumours showed a %HF of zero. The presence of hypoxia in lesions was independent of histological subtype or grade. Mean tumour K<sup>trans</sup> correlated negatively with K<sub>i</sub> (r?=?-?0.38, p?=?0.04) and %HF (r?=?-?0.33, p?=?0.04), though parametric maps exhibited intratumoural heterogeneity with hypoxic regions colocalising with both hypo- and hyperperfused areas. No correlation was observed between ADC and DCE-MRI or PET parameters. %HF correlated positively with lesion size (r?=?0.63, p?=?0.001).<h4>Conclusion</h4>Hypoxia measured by <sup>18</sup>F-FMISO-PET correlated negatively with K<sup>trans</sup> from DCE-MRI, supporting the hypothesis of perfusion-driven hypoxia in breast cancer. Intratumoural hypoxia-perfusion relationships were heterogeneous, suggesting that combined assessment may be needed for disease characterisation, which could be achieved using simultaneous multimodality imaging.<h4>Key points</h4>• At the tumour level, hypoxia measured by <sup>18</sup>F-FMISO-PET was negatively correlated with perfusion measured by DCE-MRI, which supports the hypothesis of perfusion-driven hypoxia in breast cancer. • No associations were observed between 18F-FMISO-PET parameters and tumour histology or grade, but tumour hypoxic fractions increased with lesion size. • Intratumoural hypoxia-perfusion relationships were heterogeneous, suggesting that the combined hypoxia-perfusion status of tumours may need to be considered for disease characterisation, which can be achieved via simultaneous multimodality imaging as reported here.
Project description:Paediatric brain tumours have a high mortality rate and are the most common solid tumour of childhood. Identification of high risk patients may allow for better treatment stratification. Magnetic Resonance Spectroscopy (MRS) provides a non-invasive measure of brain tumour metabolism and quantifies metabolite survival markers to aid in the clinical management of patients. Glycine can be identified using MRS and has been recently found to be important for cancer cell proliferation in tumours making it a valuable prognostic marker. The aims of this study were to investigate glycine and its added value to MRS as a prognostic marker for paediatric brain tumours in a clinical setting. 116 children with newly diagnosed brain tumours were examined with short echo-time MRS at the Birmingham Children's Hospital and followed up for five years. Survival analysis was performed using Cox regression on the entire metabolite basis set with focus on glycine and three other established survival markers for comparison: n-acetylaspartate, scyllo-inositol and lipids at 1.3 ppm. Multivariate Cox regression was used in conjunction with risk values to establish if glycine added prognostic power when combined to the established survival markers. Glycine was found to be a marker of poor prognosis in the cohort (p < 0.05) and correlated with tumour grade (p < 0.01). The addition of glycine improved the prognostic power of MRS compared to using the combination of established survival markers alone. Tumour glycine was found to improve the MRS prediction of reduced survival in paediatric brain tumours aiding the non-invasive assessment of these children.
Project description:Kidney tumours are among the most common solid tumours in children, comprising distinct subtypes differing in many aspects, including cell-of-origin, genetics, and pathology. Pre-clinical cell models capturing the disease heterogeneity are currently lacking. Here, we describe the first paediatric cancer organoid biobank. It contains tumour and matching normal kidney organoids from over 50 children with different subtypes of kidney cancer, including Wilms tumours, malignant rhabdoid tumours, renal cell carcinomas, and congenital mesoblastic nephromas. Paediatric kidney tumour organoids retain key properties of native tumours, useful for revealing patient-specific drug sensitivities. Using single cell RNA-sequencing and high resolution 3D imaging, we further demonstrate that organoid cultures derived from Wilms tumours consist of multiple different cell types, including epithelial, stromal and blastemal-like cells. Our organoid biobank captures the heterogeneity of paediatric kidney tumours, providing a representative collection of well-characterised models for basic cancer research, drug-screening and personalised medicine.