e18149 Background: Anaplastic thyroid cancer (ATC) is a rare but highly aggressive form of thyroid cancer. The FDA has approved the use of dabrafenib and trametinib (DT) in the treatment of ATC driven by a BRAF mutation (about 40-45% of all ATC’s) based on the ROAR study. Preclinical data suggests that DT alters the immune environment around ATCs, possibly making them more susceptible to immune checkpoint inhibitors like cemiplimab (Cemi), an antibody to PD-1. Based on concerns that it would be difficult to distinguish the benefit of adding Cemi to DT based on the data from DT alone, we planned a pilot study with Cemi being added to DT at the time of progression on DT alone. Methods: Single institution pilot study of patients with BRAFm ATC where Cemi was added to DT after any tumor growth on DT (RECIST progression not required). Biopsies done pre DT, post DT, pre Cemi, and post Cemi. Inclusion criteria included RECIST 1.1 measurable disease and no contraindication to immunotherapy. Steroids were allowed. Primary objective was response rate with target of at least 1 response in 12 patients. Results: Twelve patients on DT received at least one dose of Cemi. Median age was 75 years (range 51-91). 11/12 patients were male. 3 (25%) had prior RT to the neck. Patients median time on DT prior to Cemi was 4.9 months (range 2-10 months). In the 12 patients, response rate after starting Cemi was: 3 (25%) had a partial response, 6 (50%) had stable disease, and 3 (25%) had progression of disease. The median survival after starting Cemi was 6.1 months (0.9-32.4+ months) with 3 patients who are still alive as of 1/15/2025, including 2 of the 3 patients that had a partial response. Three patients (25%) were alive at 12 months after starting Cemi. Conclusions: The addition of Cemi to DT in ATC showed promising activity in a group of BRAF mutant ATC that did worse than expected on DT alone. Research biopsy analysis is pending. This data is promising enough to further explore the addition of Cemi to DT either at time of progression on DT or perhaps as initial therapy. (Study support from Regeneron and R01CA255211-01). Clinical trial information: NCT04238624 .

BACKGROUND AND PURPOSE:The choroid of the eye is a rare site for metastatic tumor spread, and as small lesions on the periphery of brain MRI studies, these choroidal metastases are often missed. To improve their detection, we aimed to use artificial intelligence to distinguish between brain MRI scans containing normal orbits and choroidal metastases. MATERIALS AND METHODS:We present a novel hierarchical deep learning framework for sequential cropping and classification on brain MR images to detect choroidal metastases. The key innovation of this approach lies in training an orbit localization network based on a YOLOv5 architecture to focus on the orbits, isolating the structures of interest and eliminating irrelevant background information. The initial subtask of localization ensures that the input to the subsequent classification network is restricted to the precise anatomic region where choroidal metastases are likely to occur. In step 1, we trained a localization network on 386 T2-weighted brain MRI axial slices from 97 patients. Using the localized orbit images from step 1, in step 2 we trained a binary classifier network with 33 normal and 33 choroidal metastasis-containing brain MRIs. To address the challenges posed by the small data set, we used a data-efficient evolutionary strategies approach, which has been shown to avoid both overfitting and underfitting in small training sets. RESULTS:Our orbit localization model identified globes with 100% accuracy and a mean average precision (mAP) of intersection over union thresholds of 0.5-0.95 [mAP(0.5:0.95)] of 0.47 on held-out testing data. Similarly, the model generalized well to our step 2 data set, which included orbits demonstrating pathologies, achieving 100% accuracy and mAP(0.5:0.95) of 0.44. mAP(0.5:0.95) appeared low because the model could not distinguish left and right orbits. Using the cropped orbits as inputs, our evolutionary strategies-trained convolutional neural network achieved a testing set area under the curve of 0.93 (95% CI, 0.83-1.03), with 100% sensitivity and 87% specificity at the optimal Youden index. CONCLUSIONS:The semiautomated pipeline from brain MRI slices to choroidal metastasis classification demonstrates the utility of a sequential localization and classification approach, and clinical relevance for identifying small, "corner-of-the-image," easily overlooked lesions. Artificial Intelligence Level of Evidence: 5B.

BACKGROUND AND PURPOSE:To evaluate whether style transfer using generative adversarial networks (GANs) can synthesize T2 FLAIR-like MR images from non-contrast CT to enhance lesion conspicuity of brain metastases. MATERIALS AND METHODS:This retrospective study analyzed 321 patients (235 with brain metastases, 86 without) whose non-contrast CT head and T2 FLAIR MR images were paired based on lesion stability and temporal proximity. Following pre-processing, CT-MR pairs were used to train GANs for CT-to-synthetic-FLAIR style transfer with 5-fold cross-validation. PatchGAN discriminator was used. 3 UNet-family generators (Attention-UNet, SUNet, UNet++) were compared using mean-absolute-error (MAE), mean-squared-error (MSE), and structural similarity index measure (SSIM). 20 synthetic MR images generated by the best performing generator were shown to 12 neuroradiologists, who rated image quality and lesion conspicuity for CT, real MR, and synthetic MR (scale 1-5★), and realism preference between real and synthetic MR. RESULTS:UNet++ produced the best image quality, with SSIM = 0.9119 ± 0.0008, MAE = 0.1232 ± 0.0022, and MSE = 0.0201 ± 0.0004 (95% CIs, all p<0.001). Mean reader-aggregated image quality ratings were 4.28 ± 0.15★ for real MR and 4.10 ± 0.14★ for synthetic MR; mean lesion conspicuity ratings were 4.67 ± 0.13★ for real MR, 4.34 ± 0.18★ for synthetic MR, and 3.16 ± 0.44★ for CT (all 95% CIs). Synthetic MR significantly improved lesion conspicuity over CT (p<0.001), with mean paired difference (MPD) 1.18 ± 0.39 and Cohen's d=2.2; while synthetic MR showed slightly reduced conspicuity compared to real MR (p<0.001), with MPD 0.33 ± 0.23 and Cohen's d=1.6 (95% CIs). Readers selected real MR as more realistic than synthetic MR in 73.3% of comparisons (p<0.001), but synthetic image quality was not significantly inferior to real MR (p=0.078). CONCLUSION:GAN-generated synthetic T2 FLAIR images from CT improve edema conspicuity of brain metastases relative to CT.

Introduction: Uveal melanomas are rare and usually detected in adults over 60 years old. We presented herein an unusual intraocular melanocytic tumor that does not fit into any known category in a child with 10-year follow-up data. The clinical findings, course, ultrasound, MRI, and pathology, including immunohistochemistry and gene profiling, of a child with pigmented intraocular masses simulating uveal melanoma are described. The tumor is provisionally called “endopapillary uveal melanocytic tumor of childhood” based on its unusual pathology and molecular biology. Comprehensive molecular testing, including MSK-IMPACT, which interrogates 468 known cancer genetic alterations, and Archer fusion testing (RNA sequence analysis) revealed no abnormalities. Case Presentation: Our patient was a 7-month-old white boy with a suspected mass in the left eye. There was no family history of melanoma. An MRI scan revealed a 1.9 × 1.7-cm well-circumscribed, heterogeneously T1 pre-contrast hyperintense mass, which demonstrated heterogeneous low signal intensity on T2-weighted imaging. A needle biopsy using a 25-gauge needle revealed a pigmented tumor, so we performed an enucleation. Upon histopathologic examination after enucleation, the tumor was densely pigmented, without significant necrosis or internal cavitation. Tumor cells grew around vascular “papillae.” Six years later, the patient is alive and shows no evidence of local or systemic recurrence. Conclusion: Our case exhibited unique features and pathology, without any of the known genetic abnormalities associated with uveal melanoma and does not match the descriptions in the existing literature. Thus, we have named these masses “endopapillary uveal melanocytic tumor of childhood.”

BACKGROUND:The outcome for pediatric patients with high-grade glioma (HGG) remains poor. Veliparib, a potent oral poly(adenosine diphosphate-ribose) polymerase (PARP) 1/2 inhibitor, enhances the activity of radiotherapy and DNA-damaging chemotherapy. METHODS:We conducted a single-arm, non-randomized phase 2 clinical trial to determine whether treatment with veliparib and radiotherapy, followed by veliparib and temozolomide, improves progression-free survival in pediatric patients with newly diagnosed HGG without H3 K27M or BRAF mutations, compared to patient-level data from historical cohorts with closely matching clinical and molecular features. Following surgical resection, newly diagnosed children with non-metastatic HGG were screened by rapid central pathology review and molecular testing. Eligible patients were enrolled on Stratum 1 (IDH wild-type) or Stratum 2 (IDH mutant). RESULTS:Both strata were closed to accrual for futility after planned interim analyses. Among the 23 eligible patients who enrolled on Stratum 1 and received protocol therapy, the 1-year event-free survival (EFS) was 23% (standard error, SE = 9%) and the 1-year overall survival (OS) was 64% (SE = 10%). Among the 14 eligible patients who enrolled on Stratum 2 and received protocol therapy, the 1-year EFS was 57% (SE = 13%) and 1-year OS was 93% (SE = 0.7%). CONCLUSIONS:Rapid central pathology review and molecular testing for eligibility were feasible. The protocol therapy including radiation, veliparib, and temozolomide was well tolerated but failed to improve outcomes compared to clinically and molecularly matched historical control cohorts treated with higher doses of alkylator chemotherapy. CLINICALTRIALS.GOV IDENTIFIER:NCT03581292 (first posted: July 10, 2018).

BACKGROUND:Median survival for patients with Diffuse Intrinsic Pontine Glioma (DIPG) is 8-12 months. METHODS:A phase 1, open label, 3 + 3 dose-escalation trial delivered radiolabeled 124I-Omburtamab, targeting B7-H3, using MR-guided stereotactic convection-enhanced delivery (CED) into the brainstem of pediatric DIPG patients. CED was performed after completion of standard-of-care external-beam radiation therapy (EBRT). Fifty children were treated and evaluable. 124I-Omburtamab activity was escalated from 0.25 to 10.0 mCi (9.25-370 MBq) and volume escalated from 0.25 mL to 10.0 mL with serial PET/MRI post administration. Safety was the primary outcome. National Cancer Institute Common Terminology Criteria for Adverse Events were assessed for 30 days following CED of 124I-Omburtamab. Secondary outcomes included overall survival and lesion-to-whole-body absorbed dose ratio. RESULTS:The maximum tolerated activity per study protocol was determined to be 6mCi (222 MBq). The overall mean (±SD) total absorbed dose in the lesion per unit injected activity was 35.2 ± 18 cGy/MBq with a high lesion-to-whole-body absorbed dose ratio averaging 816, across all activity levels. Eleven patients had treatment-related grade 3 CNS toxicities with no grade-4 or -5 CNS toxicities. Five dose-limiting toxicity events occurred. Median survival was 15.29 months from diagnosis (95% CI: 12.20-16.83 months). Survival rate estimates at 1, 2, and 3 years were 65.4% (CI 53.3-80.1%), 18.4% (CI: 10.2-33.2%), and 11.7% (CI: 5.3-25.7%), respectively. CONCLUSIONS:Administration of 124I-Omburtamab via CED is a safe treatment option for DIPG, with a maximum tolerated activity level identified. This study represents the first-in-human theranostic use of a 124I radiopharmaceutical, simultaneously, as an imaging and therapeutic agent. TRIAL REGISTRATION:NCT01502917; https://clinicaltrials.gov/study/NCT01502917.

Background:Diffuse intrinsic pontine glioma (DIPG) carries a high mortality rate and lacks effective treatment options with a median overall survival (OS) of 8-12 months. Convection-enhanced delivery (CED) has demonstrated safety in phase I trials, but efficacy is indeterminate. Evaluating anatomic patterns of relapse may aid in determining therapeutic efficacy of local CED drug delivery strategies. Methods:Sixty-three children with DIPG were retrospectively reviewed for first radiographic progression. All patients were treated using conventional external beam radiation (EBRT) and 31 were treated with CED of radiolabeled 124-iodine-omburtamab (NCT01502917). Anatomic patterns of initial progression were coded by independent neuroradiologists. OS and cumulative incidence of progression at each anatomic site were assessed in a competing risk analysis with death as a competing variable and were stratified based on CED treatment. Results:Median OS was 14.67 months for the cohort. Patients receiving CED demonstrated higher rates of progression in general, when considering progression at all anatomical sites (HR 1.79, P = .047); no significant difference was found in OS when stratified by CED treatment (P = .22). However, CED treatment was associated with significantly lower cumulative incidence of local pontine and medullary progression (HR: 0.42, P = .03; HR 0.14, P = .01, respectively) relative to non-CED-treated patients. Conclusions:Anatomically defined patterns of relapse provide evidence for locoregional control in children with DIPG treated with radioimmunotherapy administered by CED. Future CED or local surgical therapy trials can benefit from including detailed patterns of relapse as a prospective outcome.

This phase 1, dose-escalation study examined the use of the radiolabeled antibody 124I-Omburtamab delivered directly to brain-stem tumors via convection-enhanced delivery (CED). CED bypasses the blood-brain barrier by injecting the agent under the pressure of a peristaltic pump to convectively drive the therapeutic agent through the brain tissue and tumor compartment, enabling high concentrations at the target site. We evaluated 124I-Omburtamab's potential to deliver therapeutic radiation doses to diffuse intrinsic pontine glioma in children. PET and MRI were used to assess the alignment between the intended and actual distribution of the agent, with an analysis of tumor coverage and radiation absorbed doses. Methods: 124I-Omburtamab doses ranging from 9.25 to 370 MBq were administered to 36 patients. Tumor distribution volumes were derived from PET images by placing volumes of interest over lesions and overlaying them onto T2-weighted fluid-attenuated inversion recovery MRI-delineated tumor volumes. Dosimetry metrics evaluated after CED included dose-volume histograms, tumor coverage percentage, and the Dice similarity coefficient between the antibody distribution and tumor volume. A 4-quadrant scatter plot of Dice similarity versus tumor coverage was used to classify treatment variations among patients. Results: Serial PET scans showed 124I-Omburtamab localization in brain-stem lesions from 1 h to 7 d ± 1 d after dose administration. Coverage analysis revealed that 29 patients had tumor volume coverage greater than 50%, and 28 had a Dice similarity coefficient over 50%. The 4-quadrant statistical analysis-percentage of coverage versus Dice similarity coefficient-showed that 27 patients had acceptable coverage for treatment, and 4 patients experiencing suboptimal tumor coverage. Conclusion: CED of 124I-Omburtamab is a novel approach for delivering radiolabeled therapies into brain-stem tumors. Imaging enabled quantification of radiation dose coverage within the MRI-defined tumor target, highlighting the importance of precise alignment between therapeutic agent distribution and tumor volume.

BACKGROUND:The prognosis for patients with central nervous system (CNS) retinoblastoma (RB) (trilateral or stage 4b metastatic RB) treated with high-dose chemotherapy and autologous stem cell transplant (HDC-ASCT) remains poor. The impact of irradiation when administered as part of upfront therapy post HDC-ASCT on treatment outcomes and survival is unknown. METHODS:We performed a retrospective review of all patients with CNS RB (seven stage 4b, eight trilateral, one pineal lesion belonging to methylation group RB) who underwent induction chemotherapy with an intent to proceed to HDC-ASCT at two institutions. RESULTS:Twelve of 16 patients (n = 75%) achieved an objective response to induction chemotherapy, while four patients had progressive/refractory disease; two patients responded to subsequent therapy and proceeded to ASCT, and two patients did not. Seven of 14 patients who underwent HDC-ASCT, received radiotherapy as part of upfront therapy post HDC-ASCT in the form of craniospinal irradiation (CSI) (n = 3), intraventricular radioimmunotherapy (n = 3), or both CSI and intraventricular radioimmunotherapy (n = 1). The Kaplan-Meier estimate of overall survival for these patients was 62.5% at 5 years; no patients developed second malignant neoplasms within the radiation fields. For the seven patients who did not receive radiotherapy, the overall survival was 28.6% at 5 years. CONCLUSIONS:CSI (23.4 Gy) alone or in conjunction with intraventricular RIT may have clinical utility in eliminating persistent MRD post HDC-ASCT, contributing to improved disease-free survival in patients with CNS RB. This treatment strategy merits evaluation in a prospective, multicenter clinical trial for patients with CNS metastatic RB.

10040 Background: Neuroblastoma (NB) relapsing in the central nervous system (CNS), though uncommon, is historically incurable. However, a multimodality approach has improved long term survival (1). Timely detection of CNS relapse may reduce or prevent morbidity and mortality. In most centers, surveillance for NB includes whole-body MIBG scans but does not require dedicated anatomical brain imaging. At Memorial Sloan Kettering Cancer Center (MSK), head MRI at regular intervals is standard. The objective of this retrospective study was to determine the optimal imaging modality for detecting CNS relapse in NB. Methods: After MSK IRB approval, records of patients with CNS NB seen at MSK from 2004-2023 were evaluated. In most cases, relapse was diagnosed at other institutions before referral to MSK. Analyzed data included symptoms and findings on brain CT/MRI and MIBG scans. Results: Of 206 patients with MIBG-avid CNS NB, 7 were excluded because they had CNS disease at diagnosis. For the remaining 199 patients, median time to CNS relapse from diagnosis was 18 months. 132 (66%) patients had CNS relapse at a median of 9.8 months after achieving complete remission. In 67 patients with prior systemic progression, median time to CNS relapse was 10.9 months from the last relapse. Relapse was isolated to CNS in 130/199 (65%) patients. 118 (59%) patients had neurological symptoms at time of CNS disease; 81(41%) were asymptomatic, relapse being detected on surveillance scans. Multiple (>1) parenchymal lesions were noted in 74 (37%), diffuse leptomeningeal disease without parenchymal involvement in 15 (7%) and solitary lesions in 110 (55%) patients. Median diameter of the largest lesion was 2.7 (range <0.5-6.8) cm. Anatomical imaging was performed with MRI (65%), CT (14%) or both (34%). Of those undergoing both scans, CNS relapse was missed on CT in 4/67 (6%) patients. 137 patients had MIBG scans before resection of CNS relapse. All sites of CNS disease noted on MRI/CT were positive by MIBG in 46 (33%) patients. However, MIBG scan was either totally or partly negative in 69 (50%) and 22 (16%) patients, respectively. Even for lesions ≥2cm in diameter, MIBG was completely negative in 27/57 (47%). MIBG positivity did not correlate with age, size >2cm, MYCN amplification or ALK mutation status (p>0.05 for each). Lesions <1cm and infratentorially located were more likely to be negative on MIBG scan (p<0.05). Lesions in symptomatic patients, dural lesions and hemorrhagic lesions were more likely to be positive on MIBG scan (p< 0.05). Conclusions: CNS relapse is isolated to the brain in most patients. MIBG scan has poor sensitivity for the detection of CNS NB, regardless of size or location. Although CT or MRI are both effective in detecting CNS relapse, the former can miss some lesions. We recommend brain MRI for surveillance of high-risk NB for ≥2 years after initial diagnosis or last systemic relapse. 1. J. Neurooncology 97:409, 2010.

Abstract BACKGROUND Accurate molecular diagnosis of infiltrating brainstem tumors, including diffuse midline gliomas and IDH mutant astrocytomas, is essential for prognostication and optimal therapy. Surgical biopsy of the brainstem carries risk of permanent neurological deficits and may yield insufficient or non-diagnostic tissue. We hypothesized that minimally invasive “liquid biopsy” analyzing circulating tumor cell-free DNA (cfDNA) in cerebrospinal fluid (CSF) could represent an alternative molecular diagnostic modality. METHODS We performed MSK-IMPACT, a New York State Department of Health authorized hybridization capture-based next-generation panel DNA sequencing clinical assay on 44 CSF samples from 37 unique patients with brainstem tumors. Ages at time of CSF collection ranged from 1–47 years (median: 29 years). For samples in which diagnostic alterations could not be detected, we performed a series of secondary analyses for known hotspot mutations such as H3F3A, IDH1/2, and BRAF, including manual review for mutant reads below the official clinical reporting threshold, as well as droplet digital PCR (ddPCR). RESULTS Among 44 samples included in this study, 10/44 (23%) samples had mutations detected by MSK-IMPACT using standard clinical calling criteria and 14 of the remaining 34 samples (41%) had supporting evidence of a diagnostically relevant mutation based on manual review. Testing by ddPCR was performed on 11 samples (based on DNA availability), uncovering 4 additional mutations. Overall positive diagnostic yield was 64% (28/44). In biopsied patients, CSF confirmed disease-defining mutations in 80% (16/20) of CSF samples, including H3F3A K28M (9/10; 90%) and IDH1/2 (5/6; 83%). Among 24 unbiopsied patients, a disease-defining mutation was identified in 60% (15/25) of CSF samples. CONCLUSIONS Analysis of CSF cfDNA in patients with brainstem tumors has a high diagnostic yield and obviates the need for tissue biopsy in a majority of patients. A tiered testing approach using next-generation sequencing and ddPCR assays helps maximize diagnostic information and sensitivity.

Abstract T-cell Lymphoblastic Leukemia (T-cell ALL) accounts for up to 15% of all newly diagnosed cases of acute lymphoblastic leukemia in pediatric patients (0-18 years old). Patients are at risk of opportunistic infections during induction therapy with up to 2% developing blood stream infections when their absolute neutrophil count is above 1,500. We report a case of a 15-year-old female with High-Risk T-cell ALL, CNS1 (negative CNS disease), presenting with three days of headaches, photophobia, and non-neutropenic fever. She completed induction chemotherapy one week earlier with vincristine, daunorubicin, calaspargase and a 28-day course of dexamethasone. On physical exam, she had no focal deficits or changes in sensation or strength. CBC was notable for a white blood cell count: 7500, hemoglobin: 8.6, platelets: 328,000 and ANC (Absolute Neutrophil Count): 4700. The patient was also COVID positive on admission. A contrast enhanced MRI brain demonstrated multiple diffusion restricting and peripherally enhancing lesions, concerning for intracranial abscesses. Neurosurgery performed a burr hole craniotomy and aspirated 5 mL of purulent drainage, the culture of which grew Aspergillus fumigatus complex. The patient was started empirically on Voriconazole, in addition to broad spectrum antibiotic coverage pending culture results, due to excellent CNS penetration and proven superiority to amphotericin-based therapy for invasive aspergillosis. Micafungin was added due to CT chest demonstrating nodular infiltrates, suggesting primary pulmonary aspergillosis with CNS dissemination. The patient improved symptomatically without further sequelae. Patients with COVID-19 on prolonged courses of corticosteroids are at higher risk of developing fungal co-infections regardless of neutrophil count. While pulmonary aspergillosis is the most common location of a COVID-19 aspergillus coinfection, there have been documented reports of intracranial micro-abscesses. Contrast-enhanced brain MRI should be obtained in any patient with a recent history of COVID-19 presenting with new neurological symptoms, especially in a patient with a recent history of corticosteroids.

Abstract Iopofosine I 131, a radioconjugate therapy targeting tumor lipid rafts, is being assessed in CLOVER-2 [NCT05610891]–an ongoing dose-finding study evaluating iopofosine in children, adolescents, and young adults with relapsed, refractory, recurrent high-grade gliomas (HGGs) and ependymoma. Major inclusion criteria include patients aged 10-25 years, previously confirmed HGGs or ependymoma, ≥1 measurable intracranial lesion (≥10 mm in longest diameter), and performance status ≥60. Patients are assigned 1:1 to 2 treatment arms which will be assessed in parallel. Arm 1 has 2 cycles, each composed of 2 doses of 20 mCi/m2 given 14 days apart (± 1 day). Cycle 2 is given 8 weeks (± 4 days) post-initial infusion. Arm 2 patients receive 3 cycles of 10 mCi/m2 per dose. Patients in either arm are eligible for 1 additional cycle at the physician’s discretion. If after the first 10 patients, Arm 1 is not tolerated, dosing is reduced to 15 mCi/m2 per infusion. If after the first 10 patients in Arm 2 are deemed to be ineffective and Arm 1 is deemed to be tolerated, dosing is increased to 15 mCi/m2 per infusion. Final visits occur 57 days following the last cycle initiation, with quarterly follow-ups in year 1 and annually in years 2 and 3. Primary endpoints include safety, tolerability, and progression-free survival (PFS). Secondary endpoints are dosimetry, recommended phase 2 dosing determination, overall survival, antitumor and therapeutic activity, and duration of response and clinical benefit. Safety analysis uses the Wilson score method and nonparametric Wilcoxon signed-rank test. Efficacy analyses use the Response Assessment in Pediatric Neuro-Oncology criteria and the Kaplan-Meier method to detect a median PFS of 5 months compared to a historical control of 2 months. For 90% power by a 1-sided, 1-sample log-rank test: ≥10 evaluable patients must be enrolled (≤90) across ≤15 sites.

We report two neurocutaneous melanocytosis (NCM) patients who required ventriculoperitoneal shunt placement and subsequently developed intraperitoneal melanoma. These patients with NCM are at an increased risk for developing NRAS-associated melanomas in the central nervous system, which in turn may lead to symptomatic hydrocephalus requiring cerebrospinal fluid diversion. Due to the rarity of NCM, current knowledge on disease progression and appropriate management is limited. Ongoing studies aiming to better understand this condition and inform its clinical management may help to identify risk factors for developing more severe complications.

Two significant obstacles hinder the advancement of Radiology AI. The first is the challenge of overfitting, where small training data sets can result in unreliable outcomes. The second challenge is the need for more generalizability, the lack of which creates difficulties in implementing the technology across various institutions and practices. A recent innovation, deep neuroevolution (DNE), has been introduced to tackle the overfitting issue by training on small data sets and producing accurate predictions. However, the generalizability of DNE has yet to be proven. This paper strives to overcome this barrier by demonstrating that DNE can achieve satisfactory results in diverse external validation sets. The main innovation of the work is thus showing that DNE can generalize to varied outside data. Our example use case is predicting brain metastasis from neuroblastoma, emphasizing the importance of AI with limited data sets. Despite image collection and labeling advancements, rare diseases will always constrain data availability. We optimized a convolutional neural network (CNN) with DNE to demonstrate generalizability. We trained the CNN with 60 MRI images and tested it on a separate diverse collection of images from over 50 institutions. For comparison, we also trained with the more traditional stochastic gradient descent (SGD) method, with the two variants of (1) training from scratch and (2) transfer learning. Our results show that DNE demonstrates excellent generalizability with 97% accuracy on the heterogeneous testing set, while neither form of SGD could reach 60% accuracy. DNE's ability to generalize from small training sets to external and diverse testing sets suggests that it or similar approaches may play an integral role in improving the clinical performance of AI.

This is a de-identified dataset accompanying the Nature Medicine publication titled "Nivolumab plus ipilimumab in advanced salivary gland cancer: a phase 2 trial", by Vos et al. (DOI: https://doi.org/10.1038/s41591-023-02518-x). It contains the clinical features and processed data of 64 patients with salivary gland cancer, treated with nivolumab and ipilimumab immunotherapy. These data underlie results reported in the article. Raw whole-exome and RNA sequencing data generated in this study have been made publicly available at SRA under BioProject number PRJNA940989.

Abstract Intrinsic brainstem tumors arising in pediatric, adolescent, and young adult patients comprise a spectrum of entities, predominantly diffuse midline gliomas (DMG) and IDH mutant astrocytomas. Accurate molecular diagnosis is essential for prognostication and optimal therapy. Therapeutic considerations include inclusion (in IDH mutant) or exclusion (in H3K27M) of adjuvant Temozolomide post radiotherapy and use of molecular targeted therapy (IDH inhibitors). Surgical biopsy, however, is associated with increased risk of permanent neurological deficits and may yield insufficient or non-diagnostic tissue. We hypothesized that minimally-invasive “liquid biopsy” of cerebrospinal fluid (CSF) could represent a superior diagnostic modality in this patient population, employing molecular analysis of cell free DNA (cfDNA). We analyzed 44 CSF samples from 38 unique patients for recurrent driver mutations in brainstem gliomas, including H3K27M, IDH1, and IDH2. MSK-IMPACT, a NY state-authorized hybridization capture-based next-generation panel DNA sequencing assay, was used for analysis. Samples without a detectable driver mutation by MSK-IMPACT testing were further subjected to gene targeted testing by droplet digital PCR. In all, 10/44 (22.7%) samples had mutations detected by MSK-IMPACT using standard calling criteria and 13 of the remaining 34 samples (38.2%) had supporting evidence of a mutation based on manual review. Further testing by ddPCR was performed on a subset of cases (based on DNA availability) which confirmed 7 of the previous low-level mutations and uncovered 4 additional mutations. Overall, 27/44 (61.4%) of cases had detectable evidence of a mutation, and of the 23 patients without a known tissue diagnosis, a driver in the CSF was identified for 47.8% of patients (11/23). Minimally-invasive analysis of CSF cfDNA in patients with intrinsic brainstem tumors has a high diagnostic yield and may obviate the need for tissue biopsy in a majority of patients. Testing with high sensitivity assays is valuable to maximize the rate of detection.

Background: We hypothesized that radioactive iodine (RAI) can enhance the presentation of thyroid cancer immunogenic self- and neo-antigens to enhance the clinical benefit of immune checkpoint inhibitors (ICI). We conducted a phase 1 trial of RAI in combination with durvalumab (durva; anti-PD-L1) in patients (pts) with recurrent/metastatic (R/M) thyroid cancer. We report updated clinical outcomes and biological correlates performed on trial samples. Methods: Pts were required to have RECIST measurable R/M thyroid cancer with either >1 RAI-avid tumor(s) on the most recent RAI scan or one tumor with SUVmax <10 on FDG-PET. Prior therapies were allowed. Pts received durva 1500 mg IV every 4 weeks in combination with recombinant human TSH (rhTSH)-stimulated RAI (100 mCi) administered during cycle 1. Primary endpoint was safety and secondary endpoints included progression-free survival (PFS), defined as time from first durva dose until progression (PD) or death of any cause. Pre-treatment and on-treatment biopsies of the same target lesion were obtained from enrolled pts if feasible. Bulk RNA sequencing (RNAseq) was performed to assess transcriptomic characteristics of key tumor immune profiles and their correlation to PFS. Results: 11 pts enrolled; 7 pts underwent tumor biopsies. No dose-limiting toxicities or grade ≥3 adverse events related to drug were observed. BOR was 2 pts with partial responses, 7 pts with stable disease, and 2 pts with PD. Median PFS was 9.8 months with all patients eventually having PD events. Four pts had durable PFS >12 months. RNAseq analysis was performed on the tumor biopsies. Linear correlation analysis of transcriptome data from on-treatment tumors demonstrated a strong association between PFS and transcriptional scores for HLA expression (R2=0.76), MHC class I expression (R2=0.74), and NK/T cell cytolytic activity (CYT) (R2=0.69). The on-treatment tumors from pts with PFS >12 months had higher interferon-gamma (IFN-γ), HLA, MHC class I and CYT scores than pts with PFS<12 months (p<0.01). Baseline PD-L1 expression (normalized transcripts per million) was also significantly higher in pts with PFS>12 months (p=0.02). Detectable anti-TG and/or elevated anti-TPO (>1 IU/mL) autoantibody levels in pre-treatment serum samples (n=10) correlated with longer PFS on study treatment (p<0.03). RNAseq gene set enrichment analysis (GSEA) of on- vs pre-treatment samples showed durva-RAI increased thyroid autoimmunity gene sets in addition to the induction of IFN-γ and antigen presentation pathway. Conclusions: Durva-RAI has a favorable safety profile and is associated with prolonged PFS (>12 months) in a subset of pts with R/M thyroid cancer. Transcriptomic profiling reveals that durva-RAI enhancement of tumor antigen presentation and inflammation with T cell activation correlates to prolonged disease control. Enhancing pre-existing, subclinical autoimmunity against thyroid self-antigens may contribute to ICI efficacy. Further studies are needed to evaluate these hypotheses, including how RAI may contribute to ICI efficacy. Citation Format: Antoine Desilets, Winston Wong, Gnana P. Krishnamoorthy, Eric Jeffrey Sherman, Lara Dunn, Anuja Kriplani, James Vincent Fetten, Loren S. Michel, Erin McDonald, Ravinder K. Grewal, Mona Sabra, Laura Boucai, Stephanie Fish, Sofia Haque, Irina Ostrovnaya, Ronald A. Ghossein, James A. Fagin, David G. Pfister, Alan Loh Ho. Durvalumab in combination with radioactive iodine in recurrent/metastatic thyroid cancers: Update on clinical and correlative analyses [abstract]. In: Proceedings of the AACR-AHNS Head and Neck Cancer Conference: Innovating through Basic, Clinical, and Translational Research; 2023 Jul 7-8; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2023;29(18_Suppl):Abstract nr PO-072.