BACKGROUND:The Childhood Cancer Data Initiative Molecular Characterization Initiative (MCI) provides molecular testing to patients with select tumors treated at Children's Oncology Group (COG) sites, advancing our understanding of the genetic basis of pediatric cancers and their treatment. However, the frequency of actionable pharmacogenomic variants in this cohort has not yet been explored. METHODS:Germline exome sequencing data were generated by the Institute for Genomic Medicine at Nationwide Children's Hospital. Clinically actionable pharmacogenetic variants for 13 genes were based on guidelines by the Clinical Pharmacogenetics Implementation Consortium. Pharmacogenomic diplotypes were extracted using PharmCAT. The T1K computational method was used to infer HLA alleles. Genetic ancestry was estimated by iAdmix using 1000 genomes as reference. Observed variant allele frequencies were compared to gnomAD and All of Us cohorts. RESULTS:Pharmacogenomic diplotypes were extracted for 3,177 patients. Most patients had a central nervous system (CNS) tumor (69%). Genetic composition of the population was diverse with 51% European, 21% Admixed American, and 9% African. Ninety-three percent (n = 2,940) of patients had at least one actionable pharmacogenomic phenotype necessitating modification to one or more medications. We did not observe a difference across ancestral populations in the frequency of individuals carrying at least one actionable variant (p = 0.07). Variant and allele frequencies were similar to those in the gnomAD and All of Us cohorts (R2>0.99). CONCLUSION:Overall, the vast majority (93%) of patients diagnosed with pediatric cancer within the MCI had an actionable pharmacogenomic phenotype for 13 pharmacogenes evaluated.

PURPOSETo facilitate integrated multimodal data analysis, it is critical to connect data from multiple sources to address multifaceted research questions, better understand disease biology and natural history, develop new therapies, and improve existing treatments. The Childhood Cancer Data Initiative (CCDI) Participant Index, an application programming interface, aims to address this challenge by providing a digital ID mapping and matching service which collects and cross-references all known IDs associated with a participant.METHODSA variety of retrospective and prospective data collected through the CCDI Data Ecosystem equal or surpass the complexity of patient data systems in large health care organizations. The CCDI Data Ecosystem includes participant data collected under multiple protocols and at multiple sites, which often results in the same participant being associated with multiple IDs depending on the source, time, or other variables. CCDI is exploring ways to integrate diverse data types (such as genomic, proteomic, imaging, transcriptomic, clinical trial, and electronic health record data), collected over time and from different sources at the participant level, while ensuring privacy protection.RESULTSThis mapping allows researchers to access a more complete picture of a participant, even when data are collected at different time points, organizations, protocols, and consents.CONCLUSIONThis facilitates the creation of a connected data ecosystem and promotes data reuse, which, in turn, can accelerate research and improve participant outcomes.

We report the case of a 17-year-old girl with history of anaplastic large-cell lymphoma on crizotinib therapy who presented with pain and blurred vision in the left eye. Examination demonstrated nodular non-necrotizing anterior scleritis in the left eye. The patient initially improved with oral glucocorticoid treatment until she developed nodular non-necrotizing anterior scleritis in the right eye during taper of this medication. Discontinuation of crizotinib allowed the patient to taper off glucocorticoid treatment, with subsequent resolution of scleritis.

Osteosarcoma, the most common childhood bone tumor, can occur in rare cancer predisposition syndromes; however, most are sporadic with no known predisposing factors. We investigated the frequency of SMARCAL1 putative pathogenic variants in our large ongoing study of 2119 osteosarcoma patients, their relation to patient characteristics, and the population prevalence. Our analysis uncovered a higher frequency of SMARCAL1 pathogenic variants across 3 osteosarcoma patient sets (1.8%, n = 2119) than in 2625 comparably sequenced cancer-free individuals (0.3%; P < .001). Patients with SMARCAL1 pathogenic variants had statistically significantly improved overall survival compared with patients without these variants (hazard ratio [HR] = 0.36, 95% confidence interval [CI] = 0.14 to 0.96; P = .034). In the UK Biobank (469 557 exomes), there was a 33-fold increased risk of osteosarcoma in individuals with SMARCAL1 pathogenic variants. These results identify SMARCAL1 as a new osteosarcoma predisposition gene and thus warrant follow-up to identify the mechanisms by which SMARCAL1 contributes to the etiology of osteosarcoma.

PURPOSE:The International Soft Tissue Sarcoma Database Consortium is a collaboration of the North American and European pediatric oncology cooperative groups that aims to provide treatment recommendations for pediatric patients' sarcoma diagnoses. METHODS AND MATERIALS:The International Soft Tissue Sarcoma Database Consortium radiation oncology committee has developed international consensus guidelines for the use of radiation for local therapy in pediatric patients with metastatic rhabdomyosarcoma (RMS) based on grade and quality of evidence. Specifically, the guidelines address management based on disease burden, disease location, and local therapy options that focus on radiation techniques. RESULTS:Patients who present with metastatic RMS at initial diagnosis should be strongly considered for definitive therapy to the primary site and radiation to involved regional lymph nodes after neoadjuvant chemotherapy. When feasible, local treatment of all sites of disease is recommended. Evaluation of the location, size, and extent of disease and patient prognosis should be used when deciding which radiation dose and modality therapy is most appropriate for metastatic disease. CONCLUSIONS:Although the evidence is limited, these consensus guidelines highlight consensus positions regarding the best practice treatment to assist providers as they navigate treatment decisions for their pediatric patients with metastatic RMS.

PURPOSE:Euro-EWING99 study was a large, international, prospective study recruiting patients with Ewing sarcoma (EWS) between 1999 and 2015. It assessed three different clinical questions through randomized trials. We report here the characteristics and outcomes of all patients. METHODS:Patients younger than 50 years with EWS were included in the study. They received induction chemotherapy (six courses of vincristine [day 1], ifosfamide [day 1-3], doxorubicin [day 1-3], and etoposide [day 1-3; VIDE], administered every 3 weeks), local therapy (surgery/radiotherapy), and different consolidation treatments according to clinical risk group and trial.The objectives of the study were to describe the entire cohort according to the initial staging group, to describe the survival outcomes (overall survival [OS]; progression-free survival [PFS]; and local control), and to evaluate prognostic factors associated with OS and PFS. RESULTS:Three thousand three hundred ninety-five patients were included in the study, including 2,267 with a localized disease, 614 with pleuropulmonary metastases, and 514 with extrapulmonary metastases. Ninety-eight percent of patients received ≥4 neoadjuvant VIDE courses. The modalities of local treatment and consolidation therapy differed among the three staging groups. With a median follow-up of 7.2 years, PFS of the entire cohort was 60.2% and 55.4% at 3 and 5 years, respectively. OS was 72.6% and 64.6% at 3 and 5 years, respectively. In addition to metastatic status at diagnosis, main prognostic factors included patient age, tumor volume, and histologic response both for PFS and OS, independent of metastatic status. CONCLUSION:To our knowledge, this study is the largest published series of patients with EWS and may serve as a landmark paper for EWS. It confirms the major prognostic value of the complete histologic response after neoadjuvant therapy.

BACKGROUND:Taking a family history of cancer (FH) is essential for identifying individuals with heritable cancer predisposition. PROJECT:EveryChild (Children's Oncology Group [COG] trial APEC14B1), the registration and biobanking protocol of the COG, includes suggested questions about FH in first-degree relatives and personal history of genetic syndromes (GS) in pediatric oncology patients. The validity of these items is unclear; therefore, the authors assessed the data quality and face validity of the responses. METHODS:The authors analyzed case report forms regarding FH and GS of 30,157 participants (aged birth to 21 years) with newly diagnosed pediatric cancer enrolled in APEC14B1. FH and GS data were manually curated to interpret the responses and group them into categories, followed by face validity assessment-defined as the extent to which the information provided represented what it was intended to capture. RESULTS:Responses were provided for 65.7% of participants (n = 19,810), with 6.1% reporting FH (n = 1204). Of those, 97.9% (n = 1178) included sufficient free-text detail to assess face validity, although 49.4% required manual interpretation. Among FH reports, 48.3% (n = 595) were suggestive of heritable cancer risk. GS was reported in 4.3% of responders (n = 863), with 93.3% (n = 780) showing face validity after curation. Down syndrome (n = 302) and neurofibromatosis type 1 (n = 93) were the most frequently reported syndromes, with neurofibromatosis type 1 most common in patients who had central nervous system tumors. CONCLUSIONS:Despite limitations and the need for manual curation, FH and GS data collected by using proposed questions were sufficient to identify known heritable cancer patterns. These findings support questionnaire-based data collection and highlight areas for improvement.

Abstract Background: Rare pediatric cancers encompass a broad range of diagnoses, including both adult-onset carcinomas that rarely arise in children and very rare diagnoses unique to childhood. Although each individual diagnosis is exceedingly uncommon, rare tumors collectively account for ∼10% of all children with cancer. However, for many of these cancers, little is known about their genetic drivers. Methods: The Molecular Characterization Initiative (MCI), launched in 2022 by the NCI Childhood Cancer Data Initiative (CCDI), provides molecular profiling at no cost to the treating institution. Individuals 25 years old or younger newly diagnosed with a rare tumor, as defined by the Children’s Oncology Group (COG), are eligible for paired germline and tumor enhanced whole-exome sequencing and targeted RNA fusion analysis. Results are returned within 2-3 weeks from receipt of paired samples. Results: From 9/22/2022 to 03/31/2026, 872 individuals were enrolled from 160 institutions across 6 countries. The most common diagnosis subgroups were thyroid carcinoma (n=204), neuroendocrine tumors (n=92), sex cord-stromal tumors (n=75), and other rare tumors (n=182). Of the 683 patients with sequencing results, germline variants were reported in 153 patients (22.4%) and were generally consistent with established genotype-tumor associations. The most prevalent germline alterations affected DICER1 (n=37, 5.4% of total cohort), TP53 (n=16, 2.3%), RB1 (n=13,1.9%), and VHL (n=11, 1.6%). Notably, several individuals had segmental or chromosome-level copy number variants (CNVs) affecting these genes that would have been challenging to detect with smaller targeted sequencing panels. In addition, 52.3% (n=357) of tumors demonstrated somatic single-nucleotide variants (SNVs) or indels, and 54.3% (n=371) harbored somatic CNVs. Novel associations included KMT2D truncation and SMARCA4 missense variants, each seen in 4 children with nasopharyngeal carcinoma. Of the 583 patients with fusion results, oncogenic fusions were identified in 130 patients (22.8%), most commonly associated with thyroid cancer (affecting RET, NTRK3, and other kinases), desmoplastic small round cell tumor (EWSR1::WT1), or mucoepidermoid carcinoma (CRTC1::MAML2). Novel fusions included ADGRG12::NOTCH2 (adenoid cystic carcinoma), ERC1::BRAF (pancreatoblastoma), and RB1::DIAPH3 (retinoblastoma). Among the 229 patients with follow-up data, 21 (9.2%) were reported to receive a therapy matched to a molecular alteration identified by MCI. Conclusion: MCI has provided genomic profiling for children with a wide range of rare tumors across international COG centers. A reportable germline alteration of a cancer predisposition gene was identified in 22.4% of individuals with rare childhood cancers. These results highlight the importance of molecular profiling for the purposes of genetic counseling, tailoring therapy, and improving our understanding of the genetic drivers of rare pediatric tumors. Citation Format: Lauren Vasta, Jin Piao, Lea F. Surrey, John Hicks, Erin R. Rudzinski, Junne Kamihara, Jack F. Shern, Subhashini Jagu, Gregory Reaman, Malcolm Smith, Catherine Cottrell, Douglas S. Hawkins, Kris Ann P. Schultz, Theodore W. Laetsch, Kenneth S. Chen. The CCDI-COG Molecular Characterization Initiative (MCI) in rare pediatric cancers [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Breaking Barriers in the Fight against Rare Cancers; 2026 Jul 18-20; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2026;86(14_Suppl):Abstract nr PR001.

The Molecular Characterization Initiative (MCI), a key effort of the National Cancer Institute's Childhood Cancer Data Initiative (CCDI), was launched in 2022 in collaboration with the Children's Oncology Group (COG) to bring comprehensive genomic and molecular profiling to children, adolescents, and young adults diagnosed with cancer. The MCI provides paired tumor and germline molecular testing, with results returned to clinicians to inform care. Deidentified data are made available to the research community through the CCDI Data Ecosystem to facilitate the discovery of new treatment strategies. This commentary outlines the MCI's development, key accomplishments to date, and its role in laying the foundation for standardized clinical diagnostics in pediatric oncology.

Background The Children's Oncology Group (COG), formed in 2000, is a large international trial network that has conducted more than 330 clinical trials for pediatric, adolescent, and young adult cancers. The role of COG trial data for US Food and Drug Administration (FDA) approval of pediatric oncology indications was assessed since 2000. Methods All FDA approvals for pediatric oncology indications between January 1, 2000, and December 31, 2024, were reviewed. Data were collected from the FDA website, clinicaltrials.gov, and COG database. Results Twenty-five (47%) of the 53 FDA pediatric indication approvals for 37 different drugs or biologics used COG clinical trial data for the approval of the pediatric indication. The 21 COG studies used for approvals were supported by 3 partnership models, with 14 indications (26% of all pediatric approvals) supported by data from 15 trials utilizing the COG cooperative agreement award from the National Cancer Institute. Conclusion: Almost half of FDA-approved pediatric indications for oncology therapeutics were supported by data from COG trials, highlighting the effectiveness of this public-private partnership model in producing the level of clinical evidence required for regulatory success.

BACKGROUND:The Children's Oncology Group (COG), formed in 2000, is a large international trial network that has conducted over 330 clinical trials for pediatric, adolescent, and young adult cancers. The role of COG trial data for Food and Drug Administration (FDA) approval of pediatric oncology indications was assessed since 2000. METHODS:All FDA approvals for pediatric oncology indications between January 1, 2000, and December 31, 2024, were reviewed. Data were collected from the FDA website, clinicaltrials.gov and COG database. RESULTS:Twenty-five (47%) of the 53 FDA pediatric indication approvals for 37 different drugs or biologics utilized COG clinical trial data for the approval of the pediatric indication. The 21 COG studies used for approvals were supported by three different partnership models, with 14 indications (26% of all pediatric approvals) supported by data from 15 trials utilizing the COG cooperative agreement award from the National Cancer Institute (NCI). CONCLUSION:Almost half of FDA approved pediatric indications for oncology therapeutics were supported by data from COG trials, highlighting the effectiveness of this public-private partnership model in producing the level of clinical evidence required for regulatory success.

Importance Blinatumomab, a novel immunotherapy administered as a 28-day continuous infusion, has fundamentally shifted the treatment paradigm for pediatric B-cell acute lymphoblastic leukemia (B-ALL) and is now considered a component of standard therapy for the most common childhood cancer. However, the care delivery challenges in transitioning from an experimental agent on a clinical trial to widespread clinical implementation are unknown. Objective To characterize the clinical landscape of pediatric blinatumomab care-delivery practice and challenges in the US from the perspective of treating centers. Design, Setting, and Participants This survey study was conducted from February to March 2025 among US member institutions of the Children’s Oncology Group (COG) across 44 states. Exposure Institutional characteristics, including participation in the National Cancer Institute Community Oncology Research Program (NCORP), US census region, site-reported annual pediatric ALL patient volume, and prior blinatumomab experience, were assessed. Main Outcomes and Measures Incorporation of blinatumomab as standard therapy by pediatric B-ALL subtype; major outpatient care-delivery challenges defined as 4 or 5 on a 5-point Likert scale by more than 25% of centers. Results Of 195 active US COG member institutions, 147 centers completed the survey and were successfully matched with a unique COG identifier, among which 35 institutions (23.8%) were NCORP participants. There were 32 institutions (21.8%) in Midwest, 30 institutions (20.4%) in Northeast, 60 institutions (40.8%) in South, and 25 institutions (17.0%) in West census regions. Most centers reported using blinatumomab as their institutional standard therapy for National Cancer Institute standard risk–average (134 centers [91.2%]), standard risk–high (144 centers [98.0%]), and high-risk (140 centers [95.2%]) B-ALL. Fewer centers reported using blinatumomab for infant (83 centers [56.5%]) or Philadelphia chromosome–positive (96 centers [65.3%]) B-ALL. The most common major outpatient blinatumomab site care-delivery challenges included lack of home care companies (75 centers [51.0%]), family distance to treating center (41 centers [27.9%]), and insurance coverage for home care companies (37 centers [25.2%]). There were 67 sites (45.6%) that reported having no home care company options for any patients. Conclusions and Relevance In this study, challenges associated with pediatric blinatumomab home care were highly prevalent, with broader implications for health system infrastructure availability. These data highlight a need to plan for clinical implementation strategies alongside the development and testing of novel therapies.

Supplemental Description of the Rhabdomyosarcoma Web-based Application

10013 Background: The Molecular Characterization Initiative (MCI), a collaboration between the Children's Oncology Group (COG) and the NCIs Childhood Cancer Data Initiative (CCDI) which is intended to define a standardized genomic characterization of pediatric cancer, provides rapid, clinical sequencing for newly diagnosed central nervous system (CNS) tumors, rare tumors, soft tissue sarcomas (STS), or advanced stage neuroblastoma (NB) to guide diagnosis and treatment for these children. Methods: Patients enrolled on APEC14B1-MCI who are ≤25 years of age with an eligible tumor type treated at national or international COG sites with available snap frozen or FFPE tissue and paired germline samples undergo RNA/DNA extraction. Whole exome sequencing (paired tumor and germline), targeted RNA fusion (excluding NB) and methylation array (CNS clinical / STS, NB and rare tumor research only) analyses are performed, with clinical results returned in 2-3 weeks. Results: Between 3/31/22 and 11/1/24, MCI provided results for 3,972 patients, including 2,666 with CNS tumors, 781 with STS, 372 with rare tumors and 153 with NB, across 188 institutions. 89% of the > 10,000 individual tests resulted within 2 weeks of receiving nucleic acids for sequencing. Tier I/II germline single nucleotide (SNV) or copy number (CNV) variants were identified in 528 (14.1%) patients [10.4% (NB) to 25.1% (rare tumors)]. The most common germline alterations included SNVs in TP53 (n = 52,1.4%), CHEK2 (n = 50,1.3%), DICER1 (n = 35,0.93%), NF1 (n = 29,0.78%) and ATM (n = 24, 0.64%). Somatic SNVs or CNVs were identified in 85% of samples overall. Somatic SNVs most commonly involved TP53 (n = 309,8.3%), BRAF (n = 222,6.9%), or CTNNB1 (n = 166,4.4%). Targeted RNA sequencing identified gene fusions in 30% overall (23% rare, 27% CNS, 40% STS). Methylation array resulted in positive subclassification of CNS tumors in 90% of patients, including 522 patients with medulloblastoma. Additional characterization of residual nucleic acid samples is planned, with data available through the database of Genotypes and Phenotypes (dbGaP). Follow up data has been collected for 1236 patients (NCI-CCDI Hub), including frontline treatment (chemotherapy and/or radiation), response to therapy, and vital status. Additionally, 749 reported on the utility of MCI testing six months following enrollment. MCI results were used for: enrollment on a clinical trial (n = 86,11.5%), treatment with a targeted therapy (n = 8,10.7%), and/or refining the pathologic diagnosis (n = 223,29.5%). Conclusions: The MCI has resulted > 10,000 sequencing assays from 3,972 children with cancer in 31 months. This has directly impacted the diagnosis and/or management of patients with newly diagnosed tumors, providing access to timely molecular testing (including methylation in CNS tumors and fusion testing in STS), and guiding therapy and clinical trial enrollment for many patients.

Prediction of RAS pathway mutations using a trained CNN. A, Workflow for deep learning of RAS pathway mutations from FN-RMS WSIs. B and C, Representative (B) H&E images and (C) class activation maps of a RAS pathway wild-type tumor and a tumor with a KRAS p.G12C mutation (VAF = 0.659). D, Confusion matrix for predictions on a test dataset. Micro F1, Macro F1, and Matthew's correlation coefficient shown below. E, Statistics for confusion matrix. F, Average ROC curve from holdout test data.

Supplemental Figure S4. Sample partitioning for training a MYOD1 mutation predictive model using K-fold cross-validation.

PURPOSE:Precision oncology trials have generally focused on tumor testing to identify actionable alterations. The National Cancer Institute-Children's Oncology Group Pediatric MATCH trial incorporated return of germline results to assess feasibility of reporting in a cooperative group setting and characterize germline cancer predisposition in patients with refractory cancers. PATIENTS AND METHODS:Tumor and blood DNA from patients 1-21 years of age with treatment-refractory solid tumors, non-Hodgkin lymphomas, or histiocytic disorders underwent cancer gene panel sequencing. Clinical germline reports returned to 151 study sites included pathogenic/likely pathogenic (P/LP) germline variants found in 38 cancer predisposition genes (CPGs). European Society of Medical Oncology (ESMO) recommendations for germline follow-up of tumor variants in CPGs were assessed. RESULTS:Both tumor and germline reports were completed for 1,167 patients (87.5% of enrolled). A total of 295 tumor reports (25%) included 361 CPG variants of which 70 variants (19.4%) were found in the germline sample. Three additional germline-only CPG variants resulted in 73 (6.3%) of 1,167 germline reports containing variants across 21 CPGs previously associated with pediatric and/or adult cancers. Among frequently mutated CPGs in tumors, concurrent germline findings ranged from 8/32 NF1 (25.0%) and 25/163 TP53 (15.3%) to zero of 27 ALK and 18 PTEN tumor variants. ESMO guidelines recommended clinical follow-up for 110 (30.5%) of 361 tumor CPG variants which included 40 (57.1%) of 70 germline variants. CONCLUSION:Coordinated germline and tumor panel testing was feasible and revealed P/LP CPG variants in 6.3% of the Pediatric MATCH cohort. Tumor variant fraction, germline association of CPG with tumor type, and adult-oriented guidelines were not predictive of germline status, emphasizing the need for systematic germline follow-up after tumor genomic testing for pediatric patients.

10012 Background: Through collaboration with the National Cancer Institute as part of the Childhood Cancer Data Initiative, the Children’s Oncology Group offers prompt paired tissue and germline sequencing for newly diagnosed rare tumor subtypes. Methods: Individuals are eligible for paired germline and somatic blood/tissue sequencing if they are age 25 or younger, have been diagnosed with a rare tumor in the past 6 months and have both germline and tissue samples available. The paired samples undergo DNA and RNA extraction, followed by whole exome sequencing (paired tumor and germline) of cancer associated genes and RNA targeted fusion analysis. Results are returned to the primary institution within 2-3 weeks of receipt of both samples. Results: Between 09/12/2022 and 11/01/2024, 490 individuals from 123 institutions were enrolled with a total of 98 distinct diagnoses reported. The most common diagnosis groups were thyroid carcinoma (n = 120), neuroendocrine tumors (n = 53), sex cord stromal tumors (n = 41), and other carcinomas (n = 83). Of the 490 patients enrolled, 438 had submitted samples with successful return of exome results in 351/438 (80.1%) and fusion results in 302/438 (69.1%) by the data cut. Tier I/II germline single nucleotide (SNV) or copy number (CNV) variants were identified in 88 (25.1%) of patients that completed sequencing. The most prevalent germline alterations included SNVs in DICER1 (n = 20, 5.7%), TP53 (n = 7, 2.0%), RB1 (n = 7, 2.0%), CHEK2 (n = 6, 17%), SDHB (n = 5, 1.4%) and VHL (n = 5, 1.4%). 98% of samples demonstrated a Tier I/II somatic variant across 26 genes, most commonly found in DICER1 (n = 39, 11.1%), BRAF (n = 30, 8.5%), TP53 (n = 22, 6.3%) and CTNNB1 (n = 14, 4.0%). RNA fusion analysis identified positive results in 22.8% of samples. Testing identified over 33 distinct fusions. Fusions were most commonly associated with thyroid cancer; RET in 16 and NTRK in 12, or desmoplastic small round cell tumor; EWSR1::WT1 in 8. In 16.5% (n = 58) of the samples, the final diagnosis was refined based on the results of the molecular testing and 6.8% (n = 24) of the centers reported using a commercially available treatment targeting an identified molecular alternation. Conclusions: The MCI has enabled access to genetic sequencing to patients across the Children’s Oncology Group across a wide range of rare tumor diagnoses. Information about the available data can be accessed through the CCDI Hub Explore. Germline cancer predisposition was identified in a quarter of these samples, highlighting the importance of tumor-normal profiling to allow genetic counselling in these patients and appropriate surveillance. These results have the potential for lasting impact on understanding and treating individuals with rare cancers and the development of targeted future clinical trials.

Deep learning of histologic features from RMS tumor tissue. A, (i) Samples are randomly selected for training, validation, or holdout test groups with k-fold cross validation. Networks were trained to recognize (ii) basic histological characteristics and (iii) features associated with FOXO1 fusion status, or (iv) other relevant RMS mutations. (v) A predictive model was also developed to predict disease risk using only an H&E image. B, Representative H&E images (left), expert pathologist manual annotation (middle), and pixel-level segmentation results of the A.I. algorithm (right). C, Histogram of the weighted IoU scores from holdout test data (n = 29). Samples corresponding to B are indicated. D, Average and weighted intersection over union (IoU) scores of A.I. performance across a 3-fold cross-validation set compared with a pathologist manual annotation.