Identification of Multiclass Pediatric Low-Grade Neuroepithelial Tumor Molecular Subtype with ADC MR Imaging and Machine Learning

BACKGROUND AND PURPOSE

Molecular biomarker identification increasingly influences the treatment planning of pediatric low-grade neuroepithelial tumors (PLGNTs). We aimed to develop and validate a radiomics-based ADC signature predictive of the molecular status of PLGNTs.

MATERIALS AND METHODS

In this retrospective bi-institutional study, we searched the PACS for baseline brain MRIs from children with PLGNTs. Semiautomated tumor segmentation on ADC maps was performed using the semiautomated level tracing effect tool with 3D Slicer. Clinical variables, including age, sex, and tumor location, were collected from chart review. The molecular status of tumors was derived from biopsy. Multiclass random forests were used to predict the molecular status and fine-tuned using a grid search on the validation sets. Models were evaluated using independent and unseen test sets based on the combined data, and the area under the receiver operating characteristic curve (AUC) was calculated for the prediction of 3 classes: KIAA1549-BRAF fusion, BRAF V600E mutation, and non-BRAF cohorts. Experiments were repeated 100 times using different random data splits and model initializations to ensure reproducible results.

RESULTS

Two hundred ninety-nine children from the first institution and 23 children from the second institution were included (53.6% male; mean, age 8.01 years; 51.8% supratentorial; 52.2% with KIAA1549-BRAF fusion). For the 3-class prediction using radiomics features only, the average test AUC was 0.74 (95% CI, 0.73-0.75), and using clinical features only, the average test AUC was 0.67 (95% CI, 0.66-0.68). The combination of both radiomics and clinical features improved the AUC to 0.77 (95% CI, 0.75-0.77). The diagnostic performance of the per-class test AUC was higher in identifying KIAA1549-BRAF fusion tumors among the other subgroups (AUC = 0.81 for the combined radiomics and clinical features versus 0.75 and 0.74 for BRAF V600E mutation and non-BRAF, respectively).

CONCLUSIONS

ADC values of tumor segmentations have differentiative signals that can be used for training machine learning classifiers for molecular biomarker identification of PLGNTs. ADC-based pretherapeutic differentiation of the BRAF status of PLGNTs has the potential to avoid invasive tumor biopsy and enable earlier initiation of targeted therapy.

Ferumoxytol-Enhanced MRI in Children and Young Adults: State of the Art

Ferumoxytol is an ultrasmall iron oxide nanoparticle that was originally approved by the FDA in 2009 for IV treatment of iron deficiency in adults with chronic kidney disease. Subsequently, its off-label use as an MRI contrast agent increased in clinical practice, particularly in pediatric patients in North America. Unlike conventional MRI contrast agents that are based on the rare earth metal gadolinium (gadolinium-based contrast agents), ferumoxytol is biodegradable and carries no potential risk of nephrogenic systemic fibrosis. At FDA-approved doses, ferumoxytol shows no long-term tissue retention in patients with intact iron metabolism. Ferumoxytol provides unique MRI properties, including long-lasting vascular retention (facilitating high-quality vascular imaging) and retention in reticuloendothelial system tissues, thereby supporting a variety of applications beyond those possible with gadolinium-based contrast agents (GBCAs). This Clinical Perspective describes clinical and early translational applications of ferumoxytol-enhanced MRI in children and young adults through off-label use in a variety of settings, including vascular, cardiac, and cancer imaging, drawing on the institutional experience of the authors. In addition, we describe current advances in pre-clinical and clinical research using ferumoxytol in cellular and molecular imaging as well as the use of ferumoxytol as a novel potential cancer therapeutic agent.

Abstract CT001: Major tumor regressions in H3K27M-mutated diffuse midline glioma (DMG) following sequential intravenous (IV) and intracerebroventricular (ICV) delivery of GD2-CAR T cells

Background: H3K27M-mutated DMGs are universally lethal central nervous system tumors that express high levels of the disialoganglioside GD2. IV administered GD2-CAR T cells (GD2-CART) regress DMG in preclinical models, and locoregionally delivered CARs demonstrate enhanced activity in xenograft models of brain tumors.

Methods: NCT04196413 is a 3+3 Phase I dose escalation trial testing GD2-CART in patients with H3K27M DMG, with dose-limiting toxicities (DLT) considered independently for DIPG and spinal DMG (sDMG). Arm A tested escalating doses of IV GD2-CART (DL1: 1e6 GD2-CART/kg; DL2=3e6 GD2-CART/kg) following lymphodepletion (LD). After the DLT period, patients with clinical and/or radiographic benefit were eligible for subsequent ICV GD2-CART (10-30e6 GD2-CART) administered via Ommaya catheter without LD every 4-8 weeks for a maximum of 12 doses. We previously reported early results from 4 patients treated on DL1, which demonstrated clinical activity and manageable toxicity. Here we provide updated results for DL1 and DL2.

Results: Thirteen subjects were enrolled and 11 treated [n=4 DL1 (3 DIPG/1 sDMG); n=9 DL2 (7 DIPG/2 sDMG)]. Two subjects were removed prior to treatment due to rapid progression. No DLTs were observed on DL1. Three subjects experienced DLT on DL2 (2 DIPG/1 sDMG) due to grade 4 cytokine release syndrome (CRS), successfully managed with tocilizumab, anakinra, and corticosteroids. CRS occurred earlier on DL2 vs. DL1 (Day 3 vs 7). On both dose levels, all subjects exhibited transient symptoms related to on-tumor inflammation, termed Tumor Inflammation-Associated Neurotoxicity (TIAN), which was successfully managed with anakinra and, in some cases, CSF drainage and dexamethasone. No DLT due to TIAN has occurred.

Ten patients have had adequate follow-up to assess benefit. Nine experienced radiographic and/or clinical benefit after IV infusion, and they received subsequent ICV GD2-CART infusions (median= 4 ICV infusions/pt, range 1-6). ICV infusions were not associated with high-grade CRS, although some subjects developed transient fever, headache, meningismus, nausea, and/or vomiting, and several subjects developed TIAN. Four patients continue to receive ICV infusions on study and have experienced continued clinical and radiographic benefit at 11+, 9.5+, 8+ and 7+ months following enrollment. A 31-year-old with sDMG has experienced a near-complete (>95%) reduction in tumor volume and a 17-year-old with DIPG experienced a near-complete (>98%) reduction in volume of a pontine tumor.

Conclusions: IV treatment of DIPG and sDMG with GD2-CART is safe at a dose of 1e6/kg, but associated with unacceptable rates of high-grade CRS at 3e6/kg. ICV GD2-CART without LD, administered following a previous course of IV GD2-CART with LD, has been well tolerated and has mediated impressive sustained clinical benefit in some patients with DIPG/sDMG. Given these findings, we are launching a new arm to assess safety and activity and to define the recommended phase 2 dose for ICV delivery of GD2-CART without LD. Patients are eligible for up to 12 ICV infusions of GD2-CART administered every 4-6 weeks. Clinical benefit will be formally assessed using patient-reported outcomes. GD2-CART has the potential to transform therapy for patients with H3K27M+ DIPG/sDMG.

Citation Format: Robbie G. Majzner, Jasia Mahdi, Sneha Ramakrishna, Shabnum Patel, Harshini Chinnasamy, Kristen Yeom, Liora Schultz, Valentin Barsan, Rebecca Richards, Cynthia Campen, Agnes Reschke, Angus Martin Shaw Toland, Christina Baggott, Sharon Mavroukakis, Emily Egeler, Jennifer Moon, Ashley Jacobs, Karen Yamabe-Kwong, Lindsey Rasmussen, Esther Nie, Sean Green, Michael Kunicki, Michelle Fujimoto, Zach Ehlinger, Warren Reynolds, Snehit Prabhu, Katherine E. Warren, Tim Cornell, Sonia Partap, Paul Fisher, Gerald Grant, Hannes Vogel, Bita Sahaf, Kara Davis, Steven Feldman, Michelle Monje, Crystal L. Mackall. Major tumor regressions in H3K27M-mutated diffuse midline glioma (DMG) following sequential intravenous (IV) and intracerebroventricular (ICV) delivery of GD2-CAR T cells [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr CT001.

DIPG-15. Major tumor regressions in H3K27M-mutated diffuse midline glioma (DMG) following sequential intravenous (IV) and intracerebroventricular (ICV) delivery of GD2-CAR T-cells

BACKGROUND: H3K27M-mutated DMGs express high levels of the disialoganglioside GD2 and GD2-CAR T-cells (GD2-CART) regress DMG in preclinical models. METHODS: NCT04196413 is a 3 + 3 Phase I dose escalation trial testing GD2-CART in patients with biopsy-proved H3K27M DMG, with dose-limiting toxicities (DLT) considered independently for DIPG and spinal DMG (sDMG). Arm A tested escalating doses of IV GD2-CART (DL1=1e6 GD2-CART/kg; DL2=3e6 GD2-CART/kg) following lymphodepletion (LD). After the DLT period, patients with clinical and/or radiographic benefit were eligible for subsequent ICV GD2-CART infusions (10-30e6 GD2-CART) administered via Ommaya without LD. RESULTS: Twelve subjects were treated after standard radiotherapy, 7 of whom began treatment at the time of progression [n=4 DL1 (3 DIPG/1 sDMG); n=8 DL2 (6 DIPG/2 sDMG)]. No DLTs were observed on DL1. Three subjects experienced DLT on DL2 (2 DIPG/1 sDMG) due to grade-4 cytokine release syndrome (CRS). On both dose levels, all subjects exhibited transient symptoms related to on-tumor inflammation, termed Tumor Inflammation-Associated Neurotoxicity (TIAN); no DLT due to TIAN has occurred. Ten subjects experienced radiographic and/or clinical benefit after IV infusion and received subsequent ICV infusions (median=4 ICV infusions/pt, range=1-7). ICV infusions were not associated with high-grade CRS. Four patients continue to receive ICV infusions on study and have experienced continued clinical and radiographic benefit, currently 7-11 months following enrollment. Two patients (one sDMG, one DIPG) have experienced near-complete (>95%) tumor volume reduction. CONCLUSIONS: IV treatment of DIPG and sDMG with GD2-CART is safe at a dose of 1e6/kg, but associated with frequent high-grade CRS at 3e6/kg. ICV GD2-CART has been well tolerated and has mediated impressive sustained clinical benefit in some patients with DIPG/sDMG. Given these findings, we are launching a new arm to assess safety and activity and to define the recommended phase 2 dose for ICV delivery of GD2-CART without LD.

IMG-02. Improved prediction of postoperative paediatric cerebellar mutism syndrome using an artificial neural network

BACKGROUND: Postoperative paediatric cerebellar mutism syndrome (pCMS) is a common but severe complication which may arise following the resection of posterior fossa tumours in children. Two previous studies have aimed to preoperatively predict pCMS, with varying results. In this work, we examine the generalisation of these models and determine if pCMS can be predicted more accurately using an artificial neural network (ANN). METHODS: An overview of reviews was performed to identify risk factors for pCMS, and a retrospective dataset collected as per these defined risk factors from children undergoing resection of primary posterior fossa tumours. The ANN was trained on this dataset and its performance evaluated in comparison to logistic regression and other predictive indices via analysis of receiver operator characteristic curves. Area under the curve (AUC) and accuracy were calculated and compared using a Wilcoxon signed rank test, with p<0.05 considered statistically significant. RESULTS: 204 children were included, of whom 80 developed pCMS. The performance of the ANN (AUC 0.949; accuracy 90.9%) exceeded that of logistic regression (p<0.05) and both external models (p<0.001). CONCLUSION: Using an ANN, we show improved prediction of pCMS in comparison to previous models and conventional methods.

Improved prediction of postoperative paediatric cerebellar mutism syndrome using an artificial neural network

Background

Postoperative paediatric cerebellar mutism syndrome (pCMS) is a common but severe complication which may arise following the resection of posterior fossa tumours in children. Two previous studies have aimed to preoperatively predict pCMS, with varying results. In this work, we examine the generalisation of these models and determine if pCMS can be predicted more accurately using an artificial neural network (ANN).

Methods

An overview of reviews was performed to identify risk factors for pCMS, and a retrospective dataset collected as per these defined risk factors from children undergoing resection of primary posterior fossa tumours. The ANN was trained on this dataset and its performance evaluated in comparison to logistic regression and other predictive indices via analysis of receiver operator characteristic curves. Area under the curve (AUC) and accuracy were calculated and compared using a Wilcoxon signed rank test, with p<0.05 considered statistically significant.

Results

204 children were included, of whom 80 developed pCMS. The performance of the ANN (AUC 0.949; accuracy 90.9%) exceeded that of logistic regression (p<0.05) and both external models (p<0.001).

Conclusion

Using an ANN, we show improved prediction of pCMS in comparison to previous models and conventional methods.

Spatiotemporal changes in along-tract profilometry of cerebellar peduncles in cerebellar mutism syndrome

Aims

Cerebellar mutism syndrome occurs in 25% of children following resection of posterior fossa tumours. Characterised by mutism, emotional lability and cerebellar motor signs, the syndrome is usually reversible over weeks to months. Its pathophysiology remains unclear, but evidence from diffusion MRI studies has implicated damage to the superior cerebellar peduncles in the development of this condition. The objective of this study was to describe the application of automated tractography of the cerebellar peduncles to provide a high-resolution spatiotemporal profile of diffusion MRI changes in cerebellar mutism syndrome.

Method

A retrospective case-control study was performed at Lucille Packard Children’s Hospital, Stanford University. Thirty children with midline medulloblastoma (mean age ± standard deviation 8.8 ± 3.8 years) underwent volumetric T1-weighted and diffusion MRI at four timepoints over one year. Forty-nine healthy children (9.0 ± 4.2 years), scanned at a single timepoint, were included as age- and sex-matched controls. Cerebellar mutism syndrome status was determined by contemporaneous casenote review. Automated Fibre Quantification was used to segment each subject’s cerebellar peduncles (Figure 1), and fractional anisotropy was computed at 30 nodes along each tract. A non-parametric permutation-based method was used to generate a critical cluster size and p-value for by-node ANOVA group comparisons. Z-scores for patients’ fractional anisotropy at each node were calculated based on values from controls’ corresponding nodes; these were analysed using mixed ANOVA with post-hoc false discovery rate-corrected pairwise t-tests.

Results

13 patients developed cerebellar mutism syndrome. Automated fibre segmentation successfully identified the cerebellar peduncles in the majority of participants, but was more robust at follow-up timepoints (78.7% vs. 44.7% pre-operatively). Fractional anisotropy was significantly lower in the distal regions of the left superior cerebellar peduncle pre-operatively (p=0.0137) in patients compared to controls, although patients could not be distinguished pre-operatively with respect to cerebellar mutism syndrome status (Figure 2). Post-operative reductions in fractional anisotropy in children with cerebellar mutism syndrome were highly specific to the distal left superior cerebellar peduncle, and were most pronounced at follow-up timepoints (p=0.006; Figure 3). There were no significant differences in other cerebellar peduncles, either in along-tract fractional anisotropy or Z-scores, with respect to cerebellar mutism syndrome status.

Conclusion

A novel application of an automated tool to extract diffusion MRI data along the length of the cerebellar peduncles is described in a longitudinal retrospective cohort of paediatric medulloblastoma. Changes in fractional anisotropy in the cerebellar peduncles following tumour resection are described in a heretofore unprecedented level of spatiotemporal detail. In particular, children with post-operative cerebellar mutism syndrome show changes in the distal regions of the left superior cerebellar peduncle, and these changes persist up to a year post-operatively. These findings will have direct clinical implications for neurosurgeons performing resection of midline paediatric posterior fossa tumours.

Current Perspectives of Artificial Intelligence in Pediatric Neuroradiology: An Overview

Artificial Intelligence, Machine Learning, and myriad related techniques are becoming ever more commonplace throughout industry and society, and radiology is by no means an exception. It is essential for every radiologists of every subspecialty to gain familiarity and confidence with these techniques as they become increasingly incorporated into the routine practice in both academic and private practice settings. In this article, we provide a brief review of several definitions and techniques that are commonly used in AI, and in particular machine vision, and examples of how they are currently being applied to the setting of clinical neuroradiology. We then review the unique challenges that the adoption and application of faces within the subspecialty of pediatric neuroradiology, and how these obstacles may be overcome. We conclude by presenting specific examples of how AI is currently being applied within the field of pediatric neuroradiology and the potential opportunities that are available for future applications.

Listening to Mom in the NICU: effects of increased maternal speech exposure on language outcomes and white matter development in infants born very preterm

BACKGROUND

Infants born very preterm (< 32 weeks gestational age (GA)) are at risk for developmental language delays. Poor language outcomes in children born preterm have been linked to neurobiological factors, including impaired development of the brain's structural connectivity (white matter), and environmental factors, including decreased exposure to maternal speech in the neonatal intensive care unit (NICU). Interventions that enhance preterm infants' exposure to maternal speech show promise as potential strategies for improving short-term health outcomes. Intervention studies have yet to establish whether increased exposure to maternal speech in the NICU offers benefits beyond the newborn period for brain and language outcomes.

METHODS

This randomized controlled trial assesses the long-term effects of increased maternal speech exposure on structural connectivity at 12 months of age (age adjusted for prematurity (AA)) and language outcomes between 12 and 18 months of age AA. Study participants (N = 42) will include infants born very preterm (24-31 weeks 6/7 days GA). Newborns are randomly assigned to the treatment (n = 21) or standard medical care (n = 21) group. Treatment consists of increased maternal speech exposure, accomplished by playing audio recordings of each baby's own mother reading a children's book via an iPod placed in their crib/incubator. Infants in the control group have the identical iPod setup but are not played recordings. The primary outcome will be measures of expressive and receptive language skills, obtained from a parent questionnaire collected at 12-18 months AA. The secondary outcome will be measures of white matter development, including the mean diffusivity and fractional anisotropy derived from diffusion magnetic resonance imaging scans performed at around 36 weeks postmenstrual age during the infants' routine brain imaging session before hospital discharge and 12 months AA.

DISCUSSION

The proposed study is expected to establish the potential impact of increased maternal speech exposure on long-term language outcomes and white matter development in infants born very preterm. If successful, the findings of this study may help to guide NICU clinical practice for promoting language and brain development. This clinical trial has the potential to advance theoretical understanding of how early language exposure directly changes brain structure for later language learning.

TRIAL REGISTRATION

NIH Clinical Trials (ClinicalTrials.gov) NCT04193579 . Retrospectively registered on 10 December 2019.

Abstract CT031: GD2 CAR T cells mediate clinical activity and manageable toxicity in children and young adults with DIPG and H3K27M-mutated diffuse midline gliomas

Background: Diffuse intrinsic pontine glioma (DIPG) and other H3K27M-mutated diffuse midline gliomas (DMGs) are universally lethal central nervous system tumors. We previously discovered that the disialoganglioside GD2 is highly and homogenously expressed on H3K27M+ gliomas and demonstrated that GD2 CAR T cells are effective in preclinical models (Mount/Majzner et al., Nat Med, 2018).

Methods: Four subjects (3 DIPG, 1 spinal cord DMG; 4-25 yr; 1M/3F) were enrolled at DL1. Three subjects with H3K27M+ DIPG received 1e6 autologous GD2 CAR T cells/kg intravenously (IV) on study. One patient, a 25 y/o with spinal cord DMG, developed rapidly progressive disease after enrollment, resulting in complete paraparesis that led to removal from the study prior to cell infusion; she was treated on a single patient eIND with the same treatment regimen as DL1. We utilized a retroviral vector expressing a 14g2a.4-1BB.z CAR construct and an inducible iCasp9 safety switch. Manufacturing was performed in the Miltenyi Prodigy on CD4/CD8 enriched apheresis product. CAR T cells were cultured in the presence of dasatinib to improve T cell fitness (Weber et al., Science, 2021). An Ommaya reservoir was placed in all patients for monitoring of intracranial pressure (ICP).

Results: We generated GD2 CAR T cell products meeting release criteria for all four patients. All subjects received lymphodepletion with cyclophosphamide and fludarabine and remained inpatient for 14+ days after infusion. All patients developed cytokine release syndrome (Grade 1-3) manifested by fever, tachycardia and hypotension, beginning 6-7 days after infusion. Due to concern for tumoral edema and increased ICP, patients were managed with conservative fluid resuscitation, and early intervention with tocilizumab and anakinra +/- corticosteroids. Other toxicities included ICANS (Grade 1-2) and neurotoxicity mediated by inflammation in sites of disease which we have termed Tumor Inflammation-Associated Neurotoxicity (TIAN). TIAN most often manifested as worsening of existing deficits, but one patient developed symptoms of increased ICP which quickly resolved upon removal of CSF via the Ommaya. No evidence of on-target, off-tumor toxicity was observed in any patients. No dose-limiting toxicities occurred.CAR T cells trafficked to the CNS and were detected in both the CSF and peripheral blood. Inflammatory cytokines including IL-6 were elevated in the CSF and blood. 3/4 patients exhibited marked improvement or resolution of neurological deficits and some radiographic improvement. The patient treated on a single patient eIND exhibited a &gt;90% reduction in her spinal cord DMG tumor volume at two months post-infusion. Durability of the therapeutic benefit remains to be determined.

Conclusions: This is the first report of GD2 CAR T cell therapy for DIPG and spinal cord DMG. Toxicities are similar to other CAR T cells with additional, manageable complications due to inflammation at CNS sites of tumor. Treatment at DL1 demonstrated a tolerable safety profile and clear signs of T cell expansion and activity including clinical responses. This approach has the potential to transform therapy for patients with H3K27M+ DIPG/DMG. Further correlative studies, including single-cell RNAseq, longer-term outcomes and results from patients on subsequent dose levels will also be presented.

Citation Format: Robbie G. Majzner, Sneha Ramakrishna, Aaron Mochizuki, Shabnum Patel, Harshini Chinnasamy, Kristen Yeom, Liora Schultz, Rebecca Richards, Cynthia Campen, Agnes Reschke, Jasia Mahdi, Angus Martin Shaw Toland, Christina Baggott, Sharon Mavroukakis, Emily Egeler, Jennifer Moon, Kayla Landrum, Courtney Erickson, Lindsey Rasmussen, Valentin Barsan, John S. Tamaresis, Anne Cunniffe Marcy, Michael Kunicki, Michelle Fujimoto, Zach Ehlinger, Sreevidya Kurra, Timothy Cornell, Sonia Partap, Paul Fisher, Gerald Grant, Hannes Vogel, Bita Sahaf, Kara Davis, Steven Feldman, Crystal L. Mackall, Michelle Monje. GD2 CAR T cells mediate clinical activity and manageable toxicity in children and young adults with DIPG and H3K27M-mutated diffuse midline gliomas [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr CT031.

OMIC-11. SINGLE CELL RNA SEQUENCING FROM THE CSF OF SUBJECTS WITH H3K27M+ DIPG/DMG TREATED WITH GD2 CAR T-CELLULAR THERAPY

Introduction

We are conducting a Phase I clinical trial utilizing chimeric antigen receptor (CAR) T-cells targeting GD2 (NCT04196413) for H3K27M-mutant diffuse intrinsic pontine glioma (DIPG) and spinal cord diffuse midline glioma (DMG). Cerebrospinal fluid (CSF) is collected for correlative studies at the time of routine intracranial pressure monitoring via Ommaya catheter. Here we present single cell RNA-sequencing results from the first 3 subjects.

Methods

Single cell RNA-sequencing was performed utilizing 10X Genomics on cells isolated from CSF at various time points before and after CAR T-cell administration and on the CAR T-cell product. Output was aligned with Cell Ranger and analyzed in R.

Results

As detailed in the Majzner et al. abstract presented at this meeting, three of four subjects treated at dose-level one exhibited clear radiographic and/or clinical benefit. We have to date completed single cell RNA-sequencing for three of these four subjects (two with benefit, one without). After filtering out low-quality signals and doublets, 89,604 cells across 3 subjects were analyzed. Of these, 4,122 cells represent cells isolated from CSF and 85,482 cells represent CAR T-cell product. Two subjects who demonstrated clear clinical and radiographic improvement exhibited fewer S100A8⁺S100A9⁺ myeloid suppressor-cells and CD25⁺FOXP3⁺ regulatory T-cells in the CSF pre-infusion compared to the subject who did not derive a therapeutic response. In one subject with DIPG who demonstrated improvement, polyclonal CAR T-cells detectable in CSF at Day +14 demonstrated enrichment of CD8A, GZMA, GNLY and PDCD1 compared to the pre-infusion CAR T-cells by trajectory analysis, suggesting differentiation toward a cytotoxic phenotype; the same subject exhibited increasing numbers of S100A8⁺S100A9⁺ myeloid cells and CX3CR1⁺P2RY12⁺ microglia over time. Further analyses will be presented as data become available.

Conclusions

The presence of immunosuppressive myeloid populations, detectable in CSF, may correlate to clinical response in CAR T cell therapy for DIPG/DMG.

EPCT-14. GD2 CAR T-CELLS MEDIATE CLINICAL ACTIVITY AND MANAGEABLE TOXICITY IN CHILDREN AND YOUNG ADULTS WITH H3K27M-MUTATED DIPG AND SPINAL CORD DMG

Background

We previously discovered high expression of the disialoganglioside GD2 on H3K27M+ gliomas and demonstrated preclinical efficacy of intravenous (IV) GD2-targeted chimeric antigen receptor (CAR) T-cells in preclinical models of H3K27M-mutated diffuse intrinsic pontine glioma (DIPG) and diffuse midline gliomas (DMGs). We are now conducting a Phase I clinical trial (NCT04196413) of autologous GD2-targeting CAR T-cells for H3K27M+ DIPG and spinal cord DMG. Here we present the results of subjects treated at dose level 1 (DL1; 1 million GD2-CAR T-cells/kg IV).

Methods

Four patients (3 DIPG, 1 spinal DMG; ages 4–25; 1M/3F) were enrolled at DL1. Three subjects with H3K27M+ DIPG received 1e6 GD2-CAR T-cells/kg IV on study. One patient with spinal DMG enrolled but became ineligible after manufacturing and was treated on an eIND at DL1. An Ommaya reservoir was placed in all subjects for therapeutic monitoring of intracranial pressure. Subjects underwent lymphodepletion with fludarabine/cyclophosphamide and remained inpatient for at least two weeks post-infusion.

Results

All subjects developed cytokine release syndrome (Grade 1–3) manifested by fever, tachycardia and hypotension. Other toxicities included ICANS (Grade 1–2) and neurological symptoms/signs mediated by intratumoral inflammation which we have termed Tumor Inflammation-Associated Neurotoxicity (TIAN). No evidence of on-target, off-tumor toxicity was observed in any patients. No dose-limiting toxicities occurred. CAR T cells trafficked to the CNS and were detected in CSF and blood. 3/4 patients exhibited marked improvement or resolution of neurological deficits and radiographic improvement. The patient treated on an eIND exhibited &gt;90% reduction in spinal DMG volume but progressed by month 3. Re-treatment of this subject via intracerebroventricular administration resulted in a second reduction in spinal DMG volume by ~80%.

Conclusions

GD2-CAR T-cells at DL1 demonstrate a tolerable safety profile in patients with H3K27M+ DIPG/DMG with clear signs of T-cell expansion and activity including clinical responses.

Integrating neuroimaging biomarkers into the multicentre, high-dose erythropoietin for asphyxia and encephalopathy (HEAL) trial: rationale, protocol and harmonisation

INTRODUCTION

MRI and MR spectroscopy (MRS) provide early biomarkers of brain injury and treatment response in neonates with hypoxic-ischaemic encephalopathy). Still, there are challenges to incorporating neuroimaging biomarkers into multisite randomised controlled trials. In this paper, we provide the rationale for incorporating MRI and MRS biomarkers into the multisite, phase III high-dose erythropoietin for asphyxia and encephalopathy (HEAL) Trial, the MRI/S protocol and describe the strategies used for harmonisation across multiple MRI platforms.

METHODS AND ANALYSIS

Neonates with moderate or severe encephalopathy enrolled in the multisite HEAL trial undergo MRI and MRS between 96 and 144 hours of age using standardised neuroimaging protocols. MRI and MRS data are processed centrally and used to determine a brain injury score and quantitative measures of lactate and n-acetylaspartate. Harmonisation is achieved through standardisation-thereby reducing intrasite and intersite variance, real-time quality assurance monitoring and phantom scans.

ETHICS AND DISSEMINATION

IRB approval was obtained at each participating site and written consent obtained from parents prior to participation in HEAL. Additional oversight is provided by an National Institutes of Health-appointed data safety monitoring board and medical monitor.

TRIAL REGISTRATION NUMBER

NCT02811263; Pre-result.

IMG-13. MRI-BASED RADIOMICS PROGNOSTIC MARKERS OF POSTERIOR FOSSA EPENDYMOMA

PURPOSE

Posterior fossa ependymomas (PFE) are common pediatric brain tumors often assessed with MRI before surgery. Advanced radiomic analysis show promise in stratifying risk and outcome in other pediatric brain tumors. Here, we extracted high-dimensional MRI features to identify prognostic, image-based, radiomics markers of PFE and compared its performance to clinical variables.

METHODS

93 children from five centers (median age=3.3yrs; 59 males; mean PFS=50mos) were included. Tumor volumes were manually contoured on T1-post contrast and T2-weighted MRI for PyRadiomics feature extraction. Features include first-order statistics, size, shape, and texture metrics calculated on the original, log-sigma, and wavelet transformed images. Progression free survival (PFS) served as outcome. 10-fold cross-validation of a LASSO Cox regression was used to predict PFS. Model performance was analyzed and concordance metric (C) was determined using clinical variable (age at diagnosis and sex) only, radiomics only, and radiomics plus clinical variable.

RESULTS

Six radiomic features were selected (all T1): 1 first-order kurtosis (log-sigma) and 5 texture features (3 wavelet, 2 original). This model demonstrated significantly higher performance than a clinical model alone (C: 0.69 vs 0.58, p<0.001). Adding clinical features to the radiomic features didn’t improve prediction (p=0.67). For patients with molecular subtyping (n=48), adding this feature to the clinical plus radiomics models significantly improved performance over clinical features alone (C = 0.79 vs. 0.66, p=0.02). Further validation and model refinement with additional datasets are ongoing.

CONCLUSION

Our pilot study shows potential role for MRI-based radiomics and machine learning for PFE risk stratification and as radiographic biomarkers.

IMG-03. RESPONSE ASSESSMENT IN PEDIATRIC LOW-GRADE GLIOMA: RECOMMENDATIONS FROM THE RESPONSE ASSESSMENT IN PEDIATRIC NEURO-ONCOLOGY (RAPNO) WORKING GROUP

INTRODUCTION

Pediatric low-grade gliomas (pLGG) show clinical and biological features that are distinct from their adult counterparts. Consequently, additional considerations are needed for response assessment in children compared to the established adult Response Assessment in Neuro-Oncology (RANO) criteria. Standardized response criteria in pediatric clinical trials are lacking, complicating comparisons of responses across studies. We therefore established an international committee of the Radiologic Assessment in Pediatric Neuro-Oncology (RAPNO) working group to develop consensus recommendations for response assessment in pLGG.

METHODS

The committee consisted of 25 international experts in the areas of Pediatric Neuro-Oncology, Neuroradiology and Neurosurgery. The committee first developed a set of agreed upon topics they deemed necessary to understand the controversies of imaging utilization and assessment in pLGG. These topics were divided up among the committee members who presented all available literature to the entire RAPNO committee via web teleconference. Once presented, the group discussed these data and developed consensus statements and recommendations based on available literature, committee expertise and clinical experience. Each topic was discussed until a consensus was reached.

RESULTS

Final consensus included recommendations about the following topics: specific imaging sequences, advanced imaging techniques, NF1-associated pLGG, molecular and histologic classification, assessment of cysts, vision and other functional outcomes as well as overall radiologic response assessment.

CONCLUSIONS

The RAPNO pLGG consensus establishes systemic recommendations that represent an initial effort to uniformly collect and assess response in pLGG. These recommendations should now be evaluated internationally and prospectively in an effort to assess clinical utility, validate and modify as appropriate.

PATH-06. IMAGE-BASED MACHINE LEARNING CLASSIFIER FOR PEDIATRIC POSTERIOR FOSSA TUMOR HISTOPATHOLOGY

BACKGROUND

Pediatric posterior fossa (PF) tumors can include astrocytomas, ependymomas, and medulloblastomas, all of which demonstrate unique histopathology. Whole slide image analyses can be time consuming and difficult. Therefore, we used machine learning to create a screenshot-based histopathology image classifier that can distinguish between types of pediatric PF tumors.

METHODS

We took 179 histopathology slides from Stanford University, dated from 2008–2019: 87 astrocytomas, 42 ependymomas, and 50 medulloblastomas, per pathology report. Each slide was viewed under a microscope at 20x. Then, a screenshot was taken of the region of interest representative of principal slide pathology, confirmed by a trained neuropathologist. These screenshots were used to train Resnet-18 models pre-trained on the ImageNet dataset and modified to predict three classes. Various models with different hyperparameters were trained using a random hyperparameter search method. Trained models were evaluated using 5-fold cross-validation, assigning 20% of the dataset for validation with each evaluation. Qualitative analysis of model performance was assessed by creating Class Activation Map (CAM) representations of image predictions.

RESULTS

The top performing Resnet-18 model achieved a cross-validation F1 of 0.967 on categorizing screenshots of tumor pathology into three types. Qualitative analysis using CAMs indicated the model was able to identify salient distinguishing features of each tumor type.

CONCLUSIONS

We present a PF lesion classifier capable of distinguishing between astrocytomas, ependymomas, and medulloblastomas based on a histopathology screenshot. Given its ease of use, this tool has potential as an educational tool in an academic setting.

IMG-10. MRI-BASED RADIOMIC PROGNOSTIC MARKERS OF DIFFUSE MIDLINE GLIOMA

BACKGROUND

Diffuse midline gliomas (DMG) are lethal pediatric brain tumors with dismal prognoses. Presently, MRI is the mainstay of disease diagnosis and surveillance. We aimed to identify prognostic image-based radiomics markers of DMG and compare its performance to clinical variables at presentation.

METHODS

104 treatment-naïve DMG MRIs from five centers were used (median age=6.5yrs; 18 males, median OS=11mos). We isolated tumor volumes of T1-post-contrast (T1gad) and T2-weighted (T2) MRI for PyRadiomics high-dimensional feature extraction. 900 features were extracted on each image, including first order statistics, 2D/3D Shape, Gray Level Co-occurrence Matrix, Gray Level Run Length Matrix, Gray Level Size Zone Matrix, Neighboring Gray tone Difference Matrix, and Gray Level Dependence Matrix, as defined by Imaging Biomarker Standardization Initiative. Overall survival (OS) served as outcome. 10-fold cross-validation of LASSO Cox regression was used to predict OS. We analyzed model performance using clinical variable (age at diagnosis and sex) only, radiomics only, and radiomics plus clinical variable. Concordance metric was used to assess the Cox model.

RESULTS

Nine radiomic features were selected from T1gad (2 texture wavelet) and T2 (5 first-order features (1 original, 4 wavelet), 2 texture features (1 wavelet, 1 log-sigma). This model demonstrated significantly higher performance than a clinical model alone (C: 0.68 vs 0.59, p&lt;0.001). Adding clinical features to radiomic features slightly improved prediction, but was not significant (C=0.70, p=0.06).

CONCLUSION

Our pilot study shows a potential role for MRI-based radiomics and machine learning for DMG risk stratification and as image-based biomarkers for clinical therapy trials.

NFB-11. WHITE MATTER DIFFERENCES IN CHILDREN WITH NF1 COMPARED TO CONTROLS

INTRODUCTION

Neurofibromatosis type 1 (NF1) is a genetic condition in which children develop learning challenges and glioma. White matter tracts (WMT) are implicated in these cognitive functions, while oligodendroglial precursor cells are implicated in both gliomagenesis and white-matter development. Specific WMTs have not been well characterized in NF1.

METHODS

Twenty NF1 patients aged 1.4–17.6 years (M = 9.5 years, 24 male) and 20 age-and-sex-matched controls underwent dMRI at 3T (25 directions, b=1000 s/mm²). Automated segmentation of WMTs extracted fractional anisotropy (FA) and mean diffusivity (MD) of 18 major WMTs. Covariance analysis examined the effect of group (NF1/controls) on FA/MD after controlling for intracranial volume. Regression analyses for WMTs determined the interaction of FA/MD with age for NF1 patients compared to controls. Significance was set at p<0.05 after correcting for multiple comparisons using false discovery rate.

RESULTS

Compared to controls, children with NF1 had significantly decreased FA in 8 and increased MD in 12/18 tracts. Differences held after controlling for intracranial volume. The interaction between group and age accounted for a significant proportion of the variance in FA in 9 and in MD in 16/18 tracts. FA and MD differences between children with NF1 and controls were greater at younger than older ages.

CONCLUSION

Microstructural differences were observed in WMTs in children with NF1 compared to controls. These differences were not explained by intracranial volume and were most pronounced in younger children with NF1 compared to controls. These findings have implications for understanding neurocognitive deficits and gliomagenesis observed in children with NF1.

IMG-22. A DEEP LEARNING MODEL FOR AUTOMATIC POSTERIOR FOSSA PEDIATRIC BRAIN TUMOR SEGMENTATION: A MULTI-INSTITUTIONAL STUDY

BACKGROUND

Brain tumors are the most common solid malignancies in childhood, many of which develop in the posterior fossa (PF). Manual tumor measurements are frequently required to optimize registration into surgical navigation systems or for surveillance of nonresectable tumors after therapy. With recent advances in artificial intelligence (AI), automated MRI-based tumor segmentation is now feasible without requiring manual measurements. Our goal was to create a deep learning model for automated PF tumor segmentation that can register into navigation systems and provide volume output.

METHODS

720 pre-surgical MRI scans from five pediatric centers were divided into training, validation, and testing datasets. The study cohort comprised of four PF tumor types: medulloblastoma, diffuse midline glioma, ependymoma, and brainstem or cerebellar pilocytic astrocytoma. Manual segmentation of the tumors by an attending neuroradiologist served as “ground truth” labels for model training and evaluation. We used 2D Unet, an encoder-decoder convolutional neural network architecture, with a pre-trained ResNet50 encoder. We assessed ventricle segmentation accuracy on a held-out test set using Dice similarity coefficient (0–1) and compared ventricular volume calculation between manual and model-derived segmentations using linear regression.

RESULTS

Compared to the ground truth expert human segmentation, overall Dice score for model performance accuracy was 0.83 for automatic delineation of the 4 tumor types.

CONCLUSIONS

In this multi-institutional study, we present a deep learning algorithm that automatically delineates PF tumors and outputs volumetric information. Our results demonstrate applied AI that is clinically applicable, potentially augmenting radiologists, neuro-oncologists, and neurosurgeons for tumor evaluation, surveillance, and surgical planning.

NFB-04. EXAMINING DIFFUSION, ARTERIAL SPIN-LABELED PERFUSION, AND VOLUMETRIC CHANGES IN THE NEUROFIBROMATOSIS TYPE 1 BRAIN USING AN ATLAS-BASED, MULTI-PARAMETRIC APPROACH

BACKGROUND

Neurofibromatosis Type 1 (NF1) is a multisystem disorder with wide ranging clinical implications. Patients may present with macrocephaly, stroke, and cognitive deficits, all of which may impede normal neural development. We applied atlas-based, multi-parametric MRI analysis of regional brain to evaluate diffusion, arterial spin-labeled (ASL) perfusion, and volumetric changes in children with NF1.

METHODS

Children evaluated for NF1 from 2009 to 2018 at Stanford University (n=78) were retrospectively reviewed and compared to healthy controls (n=100). All patients underwent diffusion-weighted (DWI) magnetic resonance imaging at 3T, and children with brain tumors were excluded. Using atlas-based DWI analyses, we assessed volume, median apparent diffusion coefficient (ADC), and cerebral blood flow in the cerebral cortex, thalamus, caudate, putamen, globus pallidus, hippocampus, amygdala, nucleus accumbens, brain stem, and cerebral white matter. We also measured volume of the lateral ventricles. Multivariate analysis of covariance was used to test for differences between controls and NF1 patients, controlling for gender and age at time of imaging.

RESULTS

Comparing NF1 to controls, we detected increased volume and decreased ASL cerebral blood flow in white matter and all subcortical and cortical structures except for brainstem volume. Median ADC was also increased in the thalamus, pallidum, hippocampus, and brainstem.

CONCLUSIONS

Using a multi-parametric approach, we demonstrate quantitative measures of microstructural and physiologic changes of the NF1 brain. Atlas-based, quantitative MRI brain signatures may serve as biomarkers of neural development and further provide insight into associated cognitive dysfunction or risks for vasculopathy-related strokes in children with NF1.

Decoding and Systematization of Medical Imaging Features of Multiple Human Malignancies

Purpose

To summarize the data of previously reported medical imaging features on human malignancies to provide a scientific basis for more credible imaging feature selection for future studies.

Materials and Methods

A search was performed in PubMed from database inception through March 23, 2018, for studies clearly stating the decoding of medical imaging features for malignancy-related objectives and/or hypotheses. The Newcastle-Ottawa scale was used for quality assessment of the included studies. Unsupervised hierarchical clustering was performed on the manually extracted features from each included study to identify the application rules of medical imaging features across human malignancies. CT images of 1000 retrospective patients with non–small cell lung cancer were used to reveal a pattern for the value distribution of complex texture features.

Results

A total of 5026 imaging features of malignancies affecting 20 parts of the human body from 930 original articles were collated and assessed in this study. A meta-feature construct was proposed to facilitate the investigation of details of any high-dimensional complex imaging features of malignancy. A correlation atlas was constructed to clarify the general rules of applying medical imaging features to the analysis of human malignancy. Assessment of this data revealed a pattern of value distributions of the most commonly reported texture features across human malignancies. Furthermore, the significant expression of the gene mutational signature 1B across human cancer was highly consistent with the presence of the run length imaging feature across different human malignancy types.

Conclusion

The results of this study may facilitate more credible imaging feature selection in all oncology tasks across a wide spectrum of human malignancies and help to reduce bias and redundancies in future medical imaging studies.

Keywords: Computer Aided Diagnosis (CAD), Computer Applications-General (Informatics), Evidence Based Medicine, Informatics, Research Design, Statistics, Technology Assessment

Supplemental material is available for this article.

Published under a CC BY 4.0 license.

60 Precision Medicine in Developmental Pediatrics: Image-based Classification of Children with Autism Spectrum Disorder using Deep Learning

Introduction/Background

Autism spectrum disorder (ASD) is a neurodevelopmental disorder resulting in challenges with social communication, sensory differences, and repetitive and restricted patterns of behavior. ASD affects approximately 1 in 66 children in North America, with boys being affected four times more frequently than girls. Currently, diagnosis is made primarily based on clinical features and no robust biomarker for ASD diagnosis has been identified. Potential image-based biomarkers to aid ASD diagnosis may include structural properties of deep gray matter regions in the brain.

Objectives

The primary objective of this work was to investigate if children with ASD show micro- and macrostructural alterations in deep gray matter structures compared to neurotypical children, and if these biomarkers can be used for an automatic ASD classification using deep learning.

Design/Methods

Quantitative apparent diffusion coefficient (ADC) magnetic resonance imaging data was obtained from 23 boys with ASD ages 0.8 – 19.6 years (mean 7.6 years) and 39 neurotypical boys ages 0.3 – 17.75 years (mean 7.6 years). An atlas-based method was used for volumetric analysis and extraction of median ADC values for each subject within the cerebral cortex, hippocampus, thalamus, caudate, putamen, globus pallidus, amygdala, and nucleus accumbens. The extracted quantitative regional volumetric and median ADC values were then used for the development and evaluation of an automatic classification method using an artificial neural network.

Results

The classification model was evaluated using 10-fold cross validation resulting in an overall accuracy of 76%, which is considerably better than chance level (62%). Specifically, 33 neurotypical boys were correctly classified, whereas 6 neurotypical boys were incorrectly classified. For the ASD group, 14 boys were correctly classified, while 9 boys were incorrectly classified. This translates to a precision of 70% for the children with ASD and 79% for neurotypical boys.

Conclusion

To the best of our knowledge, this is the first method to classify children with ASD using micro- and macrostructural properties of deep gray matter structures in the brain. The first results of the proposed deep learning method to identify children with ASD using image-based biomarkers are promising and could serve as the platform to create a more accurate and robust deep learning model for clinical application.

Neonatal Brain Microstructure and Machine-Learning-Based Prediction of Early Language Development in Children Born Very Preterm

BACKGROUND

Very-low-birth-weight preterm infants have a higher rate of language impairments compared with children born full term. Early identification of preterm infants at risk for language delay is essential to guide early intervention at the time of optimal neuroplasticity. This study examined near-term structural brain magnetic resonance imaging (MRI) and white matter microstructure assessed on diffusion tensor imaging (DTI) in relation to early language development in children born very preterm.

METHODS

A total of 102 very-low-birth-weight neonates (birthweight≤1500g, gestational age ≤32-weeks) were recruited to participate from 2010 to 2011. Near-term structural MRI was evaluated for white matter and cerebellar abnormalities. DTI fractional anisotropy, mean diffusivity, axial diffusivity, and radial diffusivity were assessed. Language development was assessed with Bayley Scales of Infant-Toddler Development-III at 18 to 22 months adjusted age. Multivariate models with leave-one-out cross-validation and exhaustive feature selection identified three brain regions most predictive of language function. Distinct logistic regression models predicted high-risk infants, defined by language scores >1 S.D. below average.

RESULTS

Of 102 children, 92 returned for neurodevelopmental testing. Composite language score mean ± S.D. was 89.0 ± 16.0; 31 of 92 children scored <85, including 15 of 92 scoring <70, suggesting moderate-to-severe delay. Children with cerebellar asymmetry had lower receptive language subscores (P = 0.016). Infants at high risk for language impairments were predicted based on regional white matter microstructure on DTI with high accuracy (sensitivity, specificity) for composite (89%, 86%), expressive (100%, 90%), and receptive language (100%, 90%).

CONCLUSIONS

Multivariate models of near-term structural MRI and white matter microstructure on DTI may assist in identification of preterm infants at risk for language impairment, guiding early intervention.

Statistical multiscale mapping of IDH1, MGMT, and microvascular proliferation in human brain tumors from multiparametric MR and spatially-registered core biopsy

We propose a statistical multiscale mapping approach to identify microscopic and molecular heterogeneity across a tumor microenvironment using multiparametric MR (mp-MR). Twenty-nine patients underwent pre-surgical mp-MR followed by MR-guided stereotactic core biopsy. The locations of the biopsy cores were identified in the pre-surgical images using stereotactic bitmaps acquired during surgery. Feature matrices mapped the multiparametric voxel values in the vicinity of the biopsy cores to the pathologic outcome variables for each patient and logistic regression tested the individual and collective predictive power of the MR contrasts. A non-parametric weighted k-nearest neighbor classifier evaluated the feature matrices in a leave-one-out cross validation design across patients. Resulting class membership probabilities were converted to chi-square statistics to develop full-brain parametric maps, implementing Gaussian random field theory to estimate inter-voxel dependencies. Corrections for family-wise error rates were performed using Benjamini-Hochberg and random field theory, and the resulting accuracies were compared. The combination of all five image contrasts correlated with outcome (P < 10⁻⁴) for all four microscopic variables. The probabilistic mapping method using Benjamini-Hochberg generated statistically significant results (α ≤ 0.05) for three of the four dependent variables: (1) IDH1, (2) MGMT, and (3) microvascular proliferation, with an average classification accuracy of 0.984 ± 0.02 and an average classification sensitivity of 1.567% ± 0.967. The images corrected by random field theory demonstrated improved classification accuracy (0.989 ± 0.008) and classification sensitivity (5.967% ± 2.857) compared with Benjamini-Hochberg. Microscopic and molecular tumor properties can be assessed with statistical confidence across the brain from minimally-invasive, mp-MR.

Comparison of the diagnostic value of MR imaging and ophthalmoscopy for the staging of retinoblastoma

PURPOSE

To compare the diagnostic value of magnetic resonance (MR) imaging and ophthalmoscopy for staging of retinoblastoma.

METHODS

MR and ophthalmoscopic images of 36 patients who underwent enucleation were evaluated retrospectively following institutional review board approval. Histopathology being the standard of reference, the sensitivity and specificity of both diagnostic modalities were compared regarding growth pattern, iris neoangiogenesis, retinal detachment, vitreous seeds and optic nerve invasion. Data were analysed via McNemar's test.

RESULTS

Both investigations showed no significant difference in accuracy for the detection of different tumour growth patterns (P = 0.80). Vitreous seeding detection was superior by ophthalmoscopy (P < 0.001). For prelaminar optic nerve invasion, MR imaging showed similar sensitivity as ophthalmoscopy but increased specificity of 40 % (CI 0.12-0.74) vs. 20 % (0.03-0.56). MR detected optic nerve involvement past the lamina cribrosa with a sensitivity of 80 % (0.28-0.99) and a specificity of 74 % (0.55-0.88). The absence of optic nerve enhancement excluded histopathological infiltration, but the presence of optic nerve enhancement included a high number of false positives (22-24 %).

CONCLUSIONS

Ophthalmoscopy remains the method of choice for determining extent within the globe while MR imaging is useful for evaluating extraocular tumour extension. Thus, both have their own strengths and contribute uniquely to the staging of retinoblastoma.

KEY POINTS

• Ophthalmoscopy: method of choice for determining extent of retinoblastoma within the globe. • MR imaging provides optimal evaluation of extrascleral and extraocular tumour extension. • Positive enhancement of the optic nerve on MRI does not necessarily indicate involvement.

Diffusion tensor imaging (DTI) with retrospective motion correction for large-scale pediatric imaging

PURPOSE

To develop and implement a clinical DTI technique suitable for the pediatric setting that retrospectively corrects for large motion without the need for rescanning and/or reacquisition strategies, and to deliver high-quality DTI images (both in the presence and absence of large motion) using procedures that reduce image noise and artifacts.

MATERIALS AND METHODS

We implemented an in-house built generalized autocalibrating partially parallel acquisitions (GRAPPA)-accelerated diffusion tensor (DT) echo-planar imaging (EPI) sequence at 1.5T and 3T on 1600 patients between 1 month and 18 years old. To reconstruct the data, we developed a fully automated tailored reconstruction software that selects the best GRAPPA and ghost calibration weights; does 3D rigid-body realignment with importance weighting; and employs phase correction and complex averaging to lower Rician noise and reduce phase artifacts. For select cases we investigated the use of an additional volume rejection criterion and b-matrix correction for large motion.

RESULTS

The DTI image reconstruction procedures developed here were extremely robust in correcting for motion, failing on only three subjects, while providing the radiologists high-quality data for routine evaluation.

CONCLUSION

This work suggests that, apart from the rare instance of continuous motion throughout the scan, high-quality DTI brain data can be acquired using our proposed integrated sequence and reconstruction that uses a retrospective approach to motion correction. In addition, we demonstrate a substantial improvement in overall image quality by combining phase correction with complex averaging, which reduces the Rician noise that biases noisy data.