Federated learning enables big data for rare cancer boundary detection

Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.

NIMG-39. RADIOMIC ANALYSIS FOR NON-INVASIVE IN VIVO PROGNOSTIC STRATIFICATION OF DE NOVO GLIOBLASTOMA PATIENTS: A MULTI-INSTITUTIONAL EVALUATION FOR GENERALIZABILITY IN THE RESPOND CONSORTIUM

PURPOSE

Multi-parametric MRI based radiomic signatures have highlighted the promise of artificial intelligence (AI) in neuro-oncology. However, inter-institution heterogeneity hinders generalization to data from unseen clinical institutions. To this end, we formulated the ReSPOND (Radiomics Signatures for PrecisiON Diagnostics) consortium for glioblastoma. Here, we seek non-invasive generalizable radiomic signatures from routine clinically-acquired MRI for prognostic stratification of glioblastoma patients.

METHODS

We identified a retrospective cohort of 606 patients with near/gross total tumor resection ( >90%), from 13 geographically-diverse institutions. All pre-operative structural MRI scans (T1,T1-Gd,T2,T2-FLAIR) were aligned to a common anatomical atlas. An automatic algorithm segmented the whole tumors (WTs) into 3 sub-compartments, i.e., enhancing (ET), necrotic core (NC), and peritumoral T2-FLAIR signal abnormality (ED). The combination of ET+NC defines the tumor core (TC). Quantitative radiomic features were extracted to generate our AI model to stratify patients into short- (< 14mts) and long-survivors ( >14mts). The model trained on 276 patients from a single institution was independently validated on 330 unseen patients from 12 left-out institutions, using the area-under-the-receiver-operating-characteristic-curve (AUC).

RESULTS

Each feature individually offered certain (limited but reproducible) value for identifying short-survivors: 1) TC closer to lateral ventricles (AUC=0.62); 2) larger ET/brain (AUC=0.61); 3) larger TC/brain (AUC=0.59); 4) larger WT/brain (AUC=0.55); 5) larger ET/WT (AUC=0.59); 6) smaller ED/WT (AUC=0.57); 7) larger ventricle deformations (AUC=0.6). Integrating all features and age, through a multivariate AI model, resulted in higher accuracy (AUC=0.7; 95% C.I.,0.64-0.77).

CONCLUSION

Prognostic stratification using basic radiomic features is highly reproducible across diverse institutions and patient populations. Multivariate integration yields relatively more accurate and generalizable radiomic signatures, across institutions. Our results offer promise for generalizable non-invasive in vivo signatures of survival prediction in patients with glioblastoma. Extracted features from clinically-acquired imaging, renders these signatures easier for clinical translation. Large-scale evaluation could contribute to improving patient management and treatment planning.

*Indicates equal authorship.

NIMG-22. PREDICTION OF GLIOBLASTOMA CELLULAR INFILTRATION AND RECURRENCE USING MACHINE LEARNING AND MULTI-PARAMETRIC MRI ANALYSIS: RESULTS FROM THE MULTI-INSTITUTIONAL RESPOND CONSORTIUM

PURPOSE

Multi-parametric MRI and artificial intelligence (AI) methods were previously used to predict peritumoral neoplastic cell infiltration and risk of future recurrence in glioblastoma, in single-institution studies. We hypothesize that important characteristics of peritumoral tissue heterogeneity captured, engineered/selected, and quantified by these methods relate to predictions generalizable in the multi-institutional ReSPOND (Radiomics Signatures for PrecisiON Diagnostics) consortium.

METHODS

To support further development, generalization, and clinical translation of our proposed method, we trained the AI model on a retrospective cohort of 29 de novo glioblastoma patients from the Hospital of the University of Pennsylvania (UPenn) (Male/Female:20/9, age:22-78 years) followed by evaluation on a prospective multi-institutional cohort of 84 glioblastoma patients (Male/Female:51/33, age:34-89 years) from Case Western Reserve University/University Hospitals (CWRU/UH, 25), New York University (NYU, 13), Ohio State University (OSU, 13), University Hospital Río Hortega (RH, 2), and UPenn (31). Features extracted from pre-resection MRI (T1, T1-Gd, T2, T2-FLAIR, ADC) were used to build our model predicting the spatial pattern of subsequent tumor recurrence. These predictions were evaluated against regions of pathology-confirmed post-resection recurrence.

RESULTS

Our model predicted the locations that later harbored tumor recurrence with sensitivity 83%, AUC 0.83 (99% CI, 0.73-0.93), and odds ratio 7.23 (99% CI, 7.09-7.37) in the prospective cohort. Odds ratio (99% CI)/AUC(99% CI) per institute were: CWRU/UH, 7.8(7.6-8.1)/0.82(0.75-0.89); NYU, 3.5(3.3-3.6)/0.84(074-0.93); OSU, 7.9(7.6-8.3)/0.8(0.67-0.94); RH, 22.7(20-25.1)/0.94(0.27-1); UPenn, 7.1(6.8-7.3)/0.83(0.76-0.91).

CONCLUSION

This is the first study that provides relatively extensive multi-institutional validated evidence that AI can provide good predictions of peritumoral neoplastic cell infiltration and future recurrence, by dissecting the MRI signal heterogeneity in peritumoral tissue. Our analyses leveraged the unique dataset of the ReSPOND consortium, which aims to develop and evaluate AI-based biomarkers for individualized prediction and prognostication, by moving from single-institution studies to generalizable, well-validated multi-institutional predictive biomarkers.

Abstract 869: Leveraging a novel analysis pipeline for outcomes predictions from routine MRIs

There is a desperate need for approaches that allow us to use routinely gathered clinical and imaging data to make outcomes predictions for patients. The acquisition of MRI images forms a standard part of clinical care for cancer patients, especially for malignant brain tumor patients. As these digital MRI scans are performed regularly, the associated digital MRI data also grows and can form the basis for longitudinal assessment of disease and prediction of clinical outcomes. However, standard modalities are coupled with non-standard modalities in most cases (some of which is institution specific), making it difficult to organize and standardize these images for analysis. In addition, for a rare cancer such as malignant brain tumors, multi-institutional collaborations are necessary in order to have appropriate power for analyses, thereby complicating the issues around organization and standardization of images for analysis. In addition, there are likely to be discrepancies among data formats intra-institutionally. To address these shortcomings and to expedite the research process, we developed a general digital MRI analysis pipeline that is able to organize messy data, automatically preprocess, segment, extract features, and produce results that are ready for statistical analysis and discovery. Such a pipeline with modular segments for major stages in the process is beneficial to investigators who are analyzing MRI data and hope to save time from the tedium of organization and data management. With a focus on efficiency, extensibility, and reproducibility, the tools that form this pipeline also aim to increase the accessibility of complex imaging analysis to a wider audience by leveraging artificial intelligence. While pipelines exist in the literature, most are not focused on clinical MRI images of brain tumors and quickly become deprecated as software dependencies change, resulting in outdated and potentially buggy software. Our pipeline implements the latest Python version using object-oriented principles, employing PEP 8 style, and is optimized to run on modern hardware thus ensuring extensibility. The pipeline itself runs intuitively via the command line interface (CLI) that requires minimal input parameters to perform its subtasks but will have the option of accepting other parameters. We showcase features of this new MRI analysis pipeline, using malignant brain tumors as an exemplar, given their disproportionate contribution to cancer morbidity and mortality.

Citation Format: Vachan Vadmal, Kristin Waite, Jill S. Barnholtz-Sloan. Leveraging a novel analysis pipeline for outcomes predictions from routine MRIs [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 869.

MRI image analysis methods and applications: an algorithmic perspective using brain tumors as an exemplar

The use of magnetic resonance imaging (MRI) in healthcare and the emergence of radiology as a practice are both relatively new compared with the classical specialties in medicine. Having its naissance in the 1970s and later adoption in the 1980s, the use of MRI has grown exponentially, consequently engendering exciting new areas of research. One such development is the use of computational techniques to analyze MRI images much like the way a radiologist would. With the advent of affordable, powerful computing hardware and parallel developments in computer vision, MRI image analysis has also witnessed unprecedented growth. Due to the interdisciplinary and complex nature of this subfield, it is important to survey the current landscape and examine the current approaches for analysis and trend trends moving forward.

Gliomas display distinct sex-based differential methylation patterns based on molecular subtype

Background

Gliomas are the most common type of primary brain tumor and one of many cancers where males are diagnosed with greater frequency than females. However, little is known about the sex-based molecular differences in glioblastomas (GBMs) or lower grade glioma (non-GBM) subtypes. DNA methylation is an epigenetic mechanism involved in regulating gene transcription. In glioma and other cancers, hypermethylation of specific gene promoters downregulates transcription and may have a profound effect on patient outcome. The purpose of this study was to determine if sex-based methylation differences exist in different glioma subtypes.

Methods

Molecular and clinical data from glioma patients were obtained from The Cancer Genome Atlas and grouped according to tumor grade and molecular subtype (IDH1/2 mutation and 1p/19q chromosomal deletion). Sex-specific differentially methylated probes (DMPs) were identified in each subtype and further analyzed to determine if they were part of differentially methylated regions (DMRs) or associated with differentially methylated DNA transcription regulatory binding motifs.

Results

Analysis of methylation data in 4 glioma subtypes revealed unique sets of both sex-specific DMPs and DMRs in each subtype. Motif analysis based on DMP position also identified distinct sex-based sets of DNA-binding motifs that varied according to glioma subtype. Downstream targets of 2 of the GBM-specific transcription binding sites, NFAT5 and KLF6, showed differential gene expression consistent with increased methylation mediating downregulation.

Conclusion

DNA methylation differences between males and females in 4 glioma molecular subtypes suggest an important, sex-specific role for DNA methylation in epigenetic regulation of gliomagenesis.

Correlation between mutation in P53, p53 expression, cytogenetics, histologic type, and survival in patients with B-cell non-Hodgkin's lymphoma

In the biology of a cell, the central role of p53 in controlling functions such as G1/S transition (check point) and DNA damage repair, and as a trigger of apoptosis, is well established. Somatic mutations or other changes in P53 have been reported in numerous tumor types, and in some of these, they are associated with poor prognosis. In this study, we examined 237 cytogenetically characterized B-cell non-Hodgkin's lymphomas (B-NHLs) for somatic changes in P53 by Southern blot analysis, by single-strand conformation polymorphism analysis (SSCP) of exon 5 through 9, and by direct sequencing of SSCP variants to determine the frequency and types of mutations and their clinical significance. In a portion of these (173 tumors), we also studied p53 expression by immunostaining. On Southern blots, no gross change was identified in P53 and no mutation was identified in exon 9. In exons 5 through 8, 27 different mutations were identified in 25 patients (23 single-base substitutions, 3 deletions, 1 duplication). Mutations in P53 were identified in 25 of 237 tumors (10.5%), which included 1 of 45 small lymphocytic lymphomas (SLLs), 2 of 38 follicular small cleaved-cell lymphomas (FSCCs), 2 of 35 follicular mixed small cleaved-cell and large-cell lymphomas (FMxs), 1 of 4 follicular large-cell lymphomas (FLCs), 1 of 14 diffuse small cleaved-cell lymphomas (DSCCs), 2 of 17 diffuse mixed small- and large-cell lymphomas (DMxs), and 16 of 84 diffuse large-cell lymphomas (DLCCs); the difference between the histologic groups was significant (P < .01). Among mantle-cell lymphoma (MC) patients, 3 of 10 had mutations. In 16 patients, the mutation was identified in specimens obtained at diagnosis. Mutation of transition type and transversion type occurred at a relative frequency of 2:1. Thirty percent occurred at CpG dinucleotide sequences and the codon for arginine was most frequently affected. Nineteen of 99 tumors with complex cytogenetic abnormalities, but none of 69 tumors with simple cytogenetic abnormalities, had mutations (P < .001). Similarly, 11 of 25 tumors with an abnormality of 17p and 8 of 143 tumors with apparently normal 17p had mutations (P < .0001). Positive correlations were found between a mutation and p53 expression (P < .001), between missense type mutations and p53 expression (P < .005), and between 17p abnormalities and p53 expression (P < .05). Twenty-two of 49 patients without mutation and 14 of 17 patients with mutations died (P < .05), but there was no significant difference in median survival. Similarly, 21 of 26 p53 positive patients died, whereas only 1 of 24 p53-negative patients died on-study (P < .001). Among p53-negative patients, mutation (P < .01) was positively associated with a fatal outcome. These findings indicate that in B-NHL, somatic changes in P53 were present in diagnostic specimens of all histologic types, but at a higher frequency in DLC and MC tumors. P53 mutation and/or expression has a negative influence on survival, and therefore can serve as prognostic indicators. Immunostaining for p53 is an effective way to screen for P53 changes in these tumors.