A Multidisciplinary Hyper-Modeling Scheme in Personalized In Silico Oncology: Coupling Cell Kinetics with Metabolism, Signaling Networks, and Biomechanics as Plug-In Component Models of a Cancer Digital Twin.

The massive amount of human biological, imaging, and clinical data produced by multiple and diverse sources necessitates integrative modeling approaches able to summarize all this information into answers to specific clinical questions. In this paper, we present a hypermodeling scheme able to combine models of diverse cancer aspects regardless of their underlying method or scale. Describing tissue-scale cancer cell proliferation, biomechanical tumor growth, nutrient transport, genomic-scale aberrant cancer cell metabolism, and cell-signaling pathways that regulate the cellular response to therapy, the hypermodel integrates mutation, miRNA expression, imaging, and clinical data. The constituting hypomodels, as well as their orchestration and links, are described. Two specific cancer types, Wilms tumor (nephroblastoma) and non-small cell lung cancer, are addressed as proof-of-concept study cases. Personalized simulations of the actual anatomy of a patient have been conducted. The hypermodel has also been applied to predict tumor control after radiotherapy and the relationship between tumor proliferative activity and response to neoadjuvant chemotherapy. Our innovative hypermodel holds promise as a digital twin-based clinical decision support system and as the core of future in silico trial platforms, although additional retrospective adaptation and validation are necessary.

Documenting the de-identification process of clinical and imaging data for AI for health imaging projects.

Artificial intelligence (AI) is revolutionizing the field of medical imaging, holding the potential to shift medicine from a reactive "sick-care" approach to a proactive focus on healthcare and prevention. The successful development of AI in this domain relies on access to large, comprehensive, and standardized real-world datasets that accurately represent diverse populations and diseases. However, images and data are sensitive, and as such, before using them in any way the data needs to be modified to protect the privacy of the patients. This paper explores the approaches in the domain of five EU projects working on the creation of ethically compliant and GDPR-regulated European medical imaging platforms, focused on cancer-related data. It presents the individual approaches to the de-identification of imaging data, and describes the problems and the solutions adopted in each case. Further, lessons learned are provided, enabling future projects to optimally handle the problem of data de-identification. CRITICAL RELEVANCE STATEMENT: This paper presents key approaches from five flagship EU projects for the de-identification of imaging and clinical data offering valuable insights and guidelines in the domain. KEY POINTS: ΑΙ models for health imaging require access to large amounts of data. Access to large imaging datasets requires an appropriate de-identification process. This paper provides de-identification guidelines from the AI for health imaging (AI4HI) projects.

Public data homogenization for AI model development in breast cancer.

BACKGROUND: Developing trustworthy artificial intelligence (AI) models for clinical applications requires access to clinical and imaging data cohorts. Reusing of publicly available datasets has the potential to fill this gap. Specifically in the domain of breast cancer, a large archive of publicly accessible medical images along with the corresponding clinical data is available at The Cancer Imaging Archive (TCIA). However, existing datasets cannot be directly used as they are heterogeneous and cannot be effectively filtered for selecting specific image types required to develop AI models. This work focuses on the development of a homogenized dataset in the domain of breast cancer including clinical and imaging data. METHODS: Five datasets were acquired from the TCIA and were harmonized. For the clinical data harmonization, a common data model was developed and a repeatable, documented "extract-transform-load" process was defined and executed for their homogenization. Further, Digital Imaging and COmmunications in Medicine (DICOM) information was extracted from magnetic resonance imaging (MRI) data and made accessible and searchable. RESULTS: The resulting harmonized dataset includes information about 2,035 subjects with breast cancer. Further, a platform named RV-Cherry-Picker enables search over both the clinical and diagnostic imaging datasets, providing unified access, facilitating the downloading of all study imaging that correspond to specific series' characteristics (e.g., dynamic contrast-enhanced series), and reducing the burden of acquiring the appropriate set of images for the respective AI model scenario. CONCLUSIONS: RV-Cherry-Picker provides access to the largest, publicly available, homogenized, imaging/clinical dataset for breast cancer to develop AI models on top. RELEVANCE STATEMENT: We present a solution for creating merged public datasets supporting AI model development, using as an example the breast cancer domain and magnetic resonance imaging images. KEY POINTS: • The proposed platform allows unified access to the largest, homogenized public imaging dataset for breast cancer. • A methodology for the semantically enriched homogenization of public clinical data is presented. • The platform is able to make a detailed selection of breast MRI data for the development of AI models.

MI-Common Data Model: Extending Observational Medical Outcomes Partnership-Common Data Model (OMOP-CDM) for Registering Medical Imaging Metadata and Subsequent Curation Processes.

PURPOSE: The explosion of big data and artificial intelligence has rapidly increased the need for integrated, homogenized, and harmonized health data. Many common data models (CDMs) and standard vocabularies have appeared in an attempt to offer harmonized access to the available information, with Observational Medical Outcomes Partnership (OMOP)-CDM being one of the most prominent ones, allowing the standardization and harmonization of health care information. However, despite its flexibility, still capturing imaging metadata along with the corresponding clinical data continues to pose a challenge. This challenge arises from the absence of a comprehensive standard representation for image-related information and subsequent image curation processes and their interlinkage with the respective clinical information. Successful resolution of this challenge holds the potential to enable imaging and clinical data to become harmonized, quality-checked, annotated, and ready to be used in conjunction, in the development of artificial intelligence models and other data-dependent use cases. METHODS: To address this challenge, we introduce medical imaging (MI)-CDM-an extension of the OMOP-CDM specifically designed for registering medical imaging data and curation-related processes. Our modeling choices were the result of iterative numerous discussions among clinical and AI experts to enable the integration of imaging and clinical data in the context of the ProCAncer-I project, for answering a set of clinical questions across the prostate cancer's continuum. RESULTS: Our MI-CDM extension has been successfully implemented for the use case of prostate cancer for integrating imaging and curation metadata along with clinical information by using the OMOP-CDM and its oncology extension. CONCLUSION: By using our proposed terminologies and standardized attributes, we demonstrate how diverse imaging modalities can be seamlessly integrated in the future.

Multicenter DSC-MRI-Based Radiomics Predict IDH Mutation in Gliomas.

To address the current lack of dynamic susceptibility contrast magnetic resonance imaging (DSC-MRI)-based radiomics to predict isocitrate dehydrogenase (IDH) mutations in gliomas, we present a multicenter study that featured an independent exploratory set for radiomics model development and external validation using two independent cohorts. The maximum performance of the IDH mutation status prediction on the validation set had an accuracy of 0.544 (Cohen's kappa: 0.145, F1-score: 0.415, area under the curve-AUC: 0.639, sensitivity: 0.733, specificity: 0.491), which significantly improved to an accuracy of 0.706 (Cohen's kappa: 0.282, F1-score: 0.474, AUC: 0.667, sensitivity: 0.6, specificity: 0.736) when dynamic-based standardization of the images was performed prior to the radiomics. Model explainability using local interpretable model-agnostic explanations (LIME) and Shapley additive explanations (SHAP) revealed potential intuitive correlations between the IDH-wildtype increased heterogeneity and the texture complexity. These results strengthened our hypothesis that DSC-MRI radiogenomics in gliomas hold the potential to provide increased predictive performance from models that generalize well and provide understandable patterns between IDH mutation status and the extracted features toward enabling the clinical translation of radiogenomics in neuro-oncology.

Differentiating low from high-grade soft tissue sarcomas using post-processed imaging parameters derived from multiple DWI models.

PURPOSE: To investigate and histopathologically validate the role of model selection in the design of novel parametric meta-maps towards the discrimination of low from high-grade soft tissue sarcomas (STSs) using multiple Diffusion Weighted Imaging (DWI) models. METHODS: DWI data of 28 patients were quantified using the mono-exponential, bi-exponential, stretched-exponential and the diffusion kurtosis model. Akaike Weights (AW) were calculated from the corrected Akaike Information Criteria (AICc) to select the most suitable model for every pixel within the tumor volume. Pseudo-colorized classification maps were then generated to depict model suitability, hypothesizing that every single model underpins different tissue properties and cannot solely characterize the whole tumor. Single model parametric maps were turned into meta-maps using the classification map and a histological validation of the model suitability results was conducted on several subregions of different tumors. Several histogram metrics were calculated from all derived maps before and after model selection, statistical analysis was conducted using the Mann-Whitney U test, p-values were adjusted for multiple comparisons and performance of all statistically significant metrics was evaluated using the Receiver Operator Characteristic (ROC) analysis. RESULTS: The histologic analysis on several tumor subregions confirmed model suitability results on these areas. Only 3 histogram metrics, all derived from the meta-maps, were found to be statistically significant in differentiating low from high-grade STSs with an AUC higher than 89 %. CONCLUSION: Embedding model selection in the design of the diffusion parametric maps yields to histogram metrics of high discriminatory power in grading STSs.

Multiexponential T2 relaxometry of benign and malignant adipocytic tumours.

BACKGROUND: We investigated a recently proposed multiexponential (Mexp) fitting method applied to T2 relaxometry magnetic resonance imaging (MRI) data of benign and malignant adipocytic tumours and healthy subcutaneous fat. We studied the T2 distributions of the different tissue types and calculated statistical metrics to differentiate benign and malignant tumours. METHODS: Twenty-four patients with primary benign and malignant adipocytic tumours prospectively underwent 1.5-T MRI with a single-slice T2 relaxometry (Carr-Purcell-Meiboom-Gill sequence, 25 echoes) prior to surgical excision and histopathological assessment. The proposed method adaptively chooses a monoexponential or biexponential model on a voxel basis based on the adjusted R2 goodness of fit criterion. Linear regression was applied on the statistical metrics derived from the T2 distributions for the classification. RESULTS: Healthy subcutaneous fat and benign lipoma were better described by biexponential fitting with a monoexponential and biexponential prevalence of 0.0/100% and 0.2/99.8% respectively. Well-differentiated liposarcomas exhibit 17.6% monoexponential and 82.4% biexponential behaviour, while more aggressive liposarcomas show larger degree of monoexponential behaviour. The monoexponential/biexponential prevalence was 47.6/52.4% for myxoid tumours, 52.8/47.2% for poorly differentiated parts of dedifferentiated liposarcomas, and 24.9/75.1% pleomorphic liposarcomas. The percentage monoexponential or biexponential model prevalence per patient was the best classifier distinguishing between malignant and benign adipocytic tumours with a 0.81 sensitivity and a 1.00 specificity. CONCLUSIONS: Healthy adipose tissue and benign lipomas showed a pure biexponential behaviour with similar T2 distributions, while decreased adipocytic cell differentiation characterising aggressive neoplasms was associated with an increased rate of monoexponential decay curves, opening a perspective adipocytic tumour classification.

Statistical and spatial correlation between diffusion and perfusion MR imaging parameters: A study on soft tissue sarcomas.

PURPOSE: The purpose of this study was to examine the correlation of diffusion and perfusion quantitative MR parameters, on patients with malignant soft tissue tumors. In addition, we investigated the spatial agreement of hallmarks of malignancy as indicated by diffusion and perfusion biomarkers respectively. METHODS: Nonlinear least squares were used for the quantification of the DWI and DCE derived parameters for 25 patients of histologically proven soft tissue sarcoma scanned at a 1.5 T scanner. 4D data were analyzed by an in house built software implemented in Python 3.5 resulting in voxel based parametric maps based on the Intra-Voxel Incoherent Motion (IVIM), Extended Toft's (ETM) and Gamma Capillary Transit time (GCTT) models. The root mean squared error (RMSE) was also used for assessing the accuracy of the DCE fitting models. RESULTS: A good Pearson's correlation (r > 0.5) was found between micro-perfusion fraction (f-IVIM) and plasma volume (vp-GCTT). There was no significant correlation between all other possible pairs of DCE and DWI derived parameters. Following thresholding the indicators of malignancy from both imaging methods, the percentage of volume overlap between regions of high cellularity and high vascular permeability ranged from 6% to 30%. CONCLUSION: A free correlation study among all DCE and DWI derived pairs of parameters, showed a linear relationship between f-IVIM and vp-GCTT in patients with soft tissue sarcomas. DCE in conjunction with DWI MRI can provide useful information on sites of aggressive characteristics for guiding the pre-operative biopsy and for overall treatment planning.

T2, T2* and spin coupling ratio as biomarkers for the study of lipomatous tumors.

BACKGROUND: Subcutaneous fat may have variable signal intensity on T2w images depending on the choice of imaging parameters. However, fatty components within tumors have a different degree of signal dependence on the acquisition scheme. This study examined the use of T2, T2* relaxometry and spin coupling related signal changes (Spin Coupling ratio, SCr) on two different imaging protocols as clinically relevant descriptors of benign and malignant lipomatous tumors. MATERIALS AND METHODS: 20 patients with benign lipomas or liposarcomas of variable histologic grade were examined at an 1.5 T scanner with Multi Echo Spin Echo (MESE) different echo spacing (ESP) in order to produce bright fat T2w images (ESP: 13.4 ms, 25 equidistant echoes) and dark fat images (ESP: 26.8 ms with 10 equidistant echoes). T2* relaxometry acquisition comprises 4 sets of in-opposed echoes (2.4-19.2 ms, ESP: 2.4 ms) Multi Echo Gradient Echo (MEGRE) sequence. All parametric maps were calculated on a pixel basis. RESULTS: Significant differences of SCr were found for five different types of lipomatous tumors (Pairwise t-test with Bonferroni correction): lipomas, well differentiated liposarcomas, myxoid liposarcomas, pleomorphic liposarcomas and poorly differentiated liposarcomas. SCr surpassed the classification performance of T2 and T2* relaxometry. DATA CONCLUSION: A novel biomarker based on spin coupling related signal loss, SCr, is indicative of lipomatous tumor histological grading. We concluded that T2, T2* and SCr can be used for the classification of fat containing tumors, which may be important for biopsy guidance in heterogeneous masses and treatment planning.

Investigating the Role of Model-Based and Model-Free Imaging Biomarkers as Early Predictors of Neoadjuvant Breast Cancer Therapy Outcome.

Imaging biomarkers (IBs) play a critical role in the clinical management of breast cancer (BRCA) patients throughout the cancer continuum for screening, diagnosis, and therapy assessment, especially in the neoadjuvant setting. However, certain model-based IBs suffer from significant variability due to the complex workflows involved in their computation, whereas model-free IBs have not been properly studied regarding clinical outcome. In this study, IBs from 35 BRCA patients who received neoadjuvant chemotherapy (NAC) were extracted from dynamic contrast-enhanced MR imaging (DCE-MRI) data with two different approaches, a model-free approach based on pattern recognition (PR), and a model-based one using pharmacokinetic compartmental modeling. Our analysis found that both model-free and model-based biomarkers can predict pathological complete response (pCR) after the first cycle of NAC. Overall, eight biomarkers predicted the treatment response after the first cycle of NAC, with statistical significance (p-value < 0.05), and three at the baseline. The best pCR predictors at first follow-up, achieving high AUC and sensitivity and specificity more than 50%, were the hypoxic component with threshold 2 (AUC 90.4%) from the PR method, and the median value of kep (AUC 73.4%) from the model-based approach. Moreover, the 80th percentile of ve achieved the highest pCR prediction at baseline with AUC 78.5%. The results suggest that the model-free DCE-MRI IBs could be a more robust alternative to complex, model-based ones such as kep and favor the hypothesis that the PR image-derived hypoxic image component captures actual tumor hypoxia information able to predict BRCA NAC outcome.

INN: Inflated Neural Networks for IPMN Diagnosis.

Intraductal papillary mucinous neoplasm (IPMN) is a precursor to pancreatic ductal adenocarcinoma. While over half of patients are diagnosed with pancreatic cancer at a distant stage, patients who are diagnosed early enjoy a much higher 5-year survival rate of 34% compared to 3% in the former; hence, early diagnosis is key. Unique challenges in the medical imaging domain such as extremely limited annotated data sets and typically large 3D volumetric data have made it difficult for deep learning to secure a strong foothold. In this work, we construct two novel "inflated" deep network architectures, InceptINN and DenseINN, for the task of diagnosing IPMN from multisequence (T1 and T2) MRI. These networks inflate their 2D layers to 3D and bootstrap weights from their 2D counterparts (Inceptionv3 and DenseNet121 respectively) trained on ImageNet to the new 3D kernels. We also extend the inflation process by further expanding the pre-trained kernels to handle any number of input modalities and different fusion strategies. This is one of the first studies to train an end-to-end deep network on multisequence MRI for IPMN diagnosis, and shows that our proposed novel inflated network architectures are able to handle the extremely limited training data (139 MRI scans), while providing an absolute improvement of 8.76% in accuracy for diagnosing IPMN over the current state-of-the-art. Code is publicly available at https://github.com/lalonderodney/INN-Inflated-Neural-Nets.

Investigating the Correlation of Ktrans With Semi-Quantitative MRI Parameters Towards More Robust and Reproducible Perfusion Imaging Biomarkers in Three Cancer Types.

MRI Imaging biomarkers (IBs) have the potential to deliver quantitative cancer descriptors of pathophysiology for non-invasively screening, diagnosing, and monitoring cancer patients across the cancer continuum. Despite a worldwide effort to standardize IBs involving major cancer organizations, significant variability of MR-based imaging biomarker across sites still hampers their clinical translation calling for more research in the field. To this end, in the present study quantitative and semi-quantitative approaches for perfusion biomarkers are compared in MRI data from three different cancer types. In particular, Ktrans a widely used but often variable across sites candidate biomarker is compared to a semi-quantitative perfusion MRI imaging biomarker (Wash-in WIN) in patients with breast, head, and neck and soft tissue sarcoma. Our results demonstrated a linear relationship between WIN and Ktrans in all cancer patients groups when a goodness of fit (high R2) criterion for ensuring adequate data quality and accuracy is met. This consistent correlation across three different cancer types indicates that the proposed semi-quantitative perfusion MRI IB can be a simpler, more robust and reproducible alternative to Ktrans for quantitative perfusion studies in oncology.

Extending 2-D Convolutional Neural Networks to 3-D for Advancing Deep Learning Cancer Classification With Application to MRI Liver Tumor Differentiation.

Deep learning (DL) architectures have opened new horizons in medical image analysis attaining unprecedented performance in tasks such as tissue classification and segmentation as well as prediction of several clinical outcomes. In this paper, we propose and evaluate a novel three-dimensional (3-D) convolutional neural network (CNN) designed for tissue classification in medical imaging and applied for discriminating between primary and metastatic liver tumors from diffusion weighted MRI (DW-MRI) data. The proposed network consists of four consecutive strided 3-D convolutional layers with 3 × 3 × 3 kernel size and rectified linear unit (ReLU) as activation function, followed by a fully connected layer with 2048 neurons and a Softmax layer for binary classification. A dataset comprising 130 DW-MRI scans was used for the training and validation of the network. To the best of our knowledge this is the first DL solution for the specific clinical problem and the first 3-D CNN for cancer classification operating directly on whole 3-D tomographic data without the need of any preprocessing step such as region cropping, annotating, or detecting regions of interest. The classification performance results, 83% (3-D) versus 69.6% and 65.2% (2-D), demonstrated significant tissue classification accuracy improvement compared to two 2-D CNNs of different architectures also designed for the specific clinical problem with the same dataset. These results suggest that the proposed 3-D CNN architecture can bring significant benefit in DW-MRI liver discrimination and potentially, in numerous other tissue classification problems based on tomographic data, especially in size-limited, disease-specific clinical datasets.

Correction: Diffusion weighted imaging in patients with rectal cancer: Comparison between Gaussian and non-Gaussian models.

[This corrects the article DOI: 10.1371/journal.pone.0184197.].

Correction: Diffusion weighted imaging in patients with rectal cancer: Comparison between Gaussian and non-Gaussian models.

[This corrects the article DOI: 10.1371/journal.pone.0184197.].

Diffusion weighted imaging in patients with rectal cancer: Comparison between Gaussian and non-Gaussian models.

PURPOSE: The purpose of this study was to compare the performance of four diffusion models, including mono and bi-exponential both Gaussian and non-Gaussian models, in diffusion weighted imaging of rectal cancer. MATERIAL AND METHODS: Nineteen patients with rectal adenocarcinoma underwent MRI examination of the rectum before chemoradiation therapy including a 7 b-value diffusion sequence (0, 25, 50, 100, 500, 1000 and 2000 s/mm2) at a 1.5T scanner. Four different diffusion models including mono- and bi-exponential Gaussian (MG and BG) and non-Gaussian (MNG and BNG) were applied on whole tumor volumes of interest. Two different statistical criteria were recruited to assess their fitting performance, including the adjusted-R2 and Root Mean Square Error (RMSE). To decide which model better characterizes rectal cancer, model selection was relied on Akaike Information Criteria (AIC) and F-ratio. RESULTS: All candidate models achieved a good fitting performance with the two most complex models, the BG and the BNG, exhibiting the best fitting performance. However, both criteria for model selection indicated that the MG model performed better than any other model. In particular, using AIC Weights and F-ratio, the pixel-based analysis demonstrated that tumor areas better described by the simplest MG model in an average area of 53% and 33%, respectively. Non-Gaussian behavior was illustrated in an average area of 37% according to the F-ratio, and 7% using AIC Weights. However, the distributions of the pixels best fitted by each of the four models suggest that MG failed to perform better than any other model in all patients, and the overall tumor area. CONCLUSION: No single diffusion model evaluated herein could accurately describe rectal tumours. These findings probably can be explained on the basis of increased tumour heterogeneity, where areas with high vascularity could be fitted better with bi-exponential models, and areas with necrosis would mostly follow mono-exponential behavior.

Apparent Diffusion Coefficient Quantification in Determining the Histological Diagnosis of Malignant Liver Lesions.

PURPOSE: Diffusion Weighted Imaging is an established diagnostic tool for accurate differential diagnosis between benign and malignant liver lesions. The aim of our study was to evaluate the role of Histogram Analysis of ADC quantification in determining the histological diagnosis as well as the grade of malignant liver tumours. To our knowledge, there is no study evaluating the role of Histogram Analysis of ADC quantification in determining the histological diagnosis as well as the grade of malignant liver tumours. METHODS: During five years, 115 patients with known liver lesions underwent Diffusion Weighted Imaging in 3Tesla MR scanner prior to core needle biopsy. Histogram analyses of ADC in regions of interest were drawn and were correlated with biopsy histological diagnosis and grading. RESULTS: Histogram analysis of ADC values shows that 5th and 30th percentile parameters have statistically significant potency of discrimination between primary and secondary lesions groups (p values 0.0036 and 0.0125 respectively). Skewness of the histogram can help discriminate between good and poor differentiated (p value 0.17). Discrimination between primary malignancy site in metastases failed for the present number of patients in each subgroup. CONCLUSION: Statistical parameters reflecting the shape of the left side of the ADC histogram can be useful for discriminating between primary and secondary lesions and also between well differentiated versus moderate or poor. For the secondary malignancies, they failed to predict the original site of tumour.

Whole-liver diffusion-weighted MRI histogram analysis: effect of the presence of colorectal hepatic metastases on the remaining liver parenchyma.

OBJECTIVES: To explore whether whole-liver diffusion-weighted MRI analysis (of the apparently normal liver parenchyma) can help differentiate between patients with colorectal liver metastasis and controls without liver disease. MATERIALS AND METHODS: Ten patients with colorectal liver metastasis and 10 controls with no focal/diffuse liver disease underwent liver MRI at 1.5 T including diffusion-weighted imaging (DWI; b-values 0, 50, 100, 500, 750, 1000). Apparent diffusion coefficient (ADC) maps were calculated from the DWI images to carry out quantitative diffusion analyses. An experienced reader performed segmentation of the apparently nondiseased liver (excluding metastases/focal liver lesions) on the ADC maps. Histogram ADC parameters were calculated and compared between the patients and the controls. RESULTS: The mean liver ADC was 0.95×10⁻³ mm²/s for the patients versus 1.03×10⁻³ mm²/s for the controls (P=0.42). The fifth percentile of the ADC was significantly lower for the patients compared with the controls (0.45 vs. 0.69 10⁻³ mm²/s, P=0.01). The SD was significantly higher in the patient group (0.30 vs. 0.22, P<0.001). Median, skewness, kurtosis, and 30th-95th percentile were not significantly different between the two groups. Areas under the receiver operator characteristics curves to differentiate patients with metastatic liver involvement from healthy controls without liver disease were 0.79 for the fifth percentile and 0.95 for the SD. CONCLUSION: Whole-liver diffusion-weighted MRI histogram analysis showed a significant shift towards lower fifth percentile ADC values and higher SD in patients with colorectal liver metastasis compared with controls without liver disease.