Acute Lymphoblastic Leukemia (ALL) is the most frequent hematologic malignancy in children and adolescents. A strong prognostic factor in ALL is given by the Minimal Residual Disease (MRD), which is a measure for the number of leukemic cells persistent in a patient. Manual MRD assessment from Multiparameter Flow Cytometry (FCM) data after treatment is time-consuming and subjective. In this work, we present an automated method to compute the MRD value directly from FCM data. We present a novel neural network approach based on the transformer architecture that learns to directly identify blast cells in a sample. We train our method in a supervised manner and evaluate it on publicly available ALL FCM data from three different clinical centers. Our method reaches a median F1 score of ≈0.94 when evaluated on 519 B-ALL samples and shows better results than existing methods on 4 different datasets.

Decisions of automated systems in healthcare can have far-reaching consequences such as delayed or incorrect treatment and thus must be explainable and comprehensible for medical experts. This also applies to the field of automated Flow Cytometry (FCM) data analysis. In leukemic cancer therapy, FCM samples are obtained from the patient's bone marrow to determine the number of remaining leukemic cells. In a manual process, called gating, medical experts draw several polygons among different cell populations on 2D plots in order to hierarchically sub-select and track down cancer cell populations in an FCM sample. Several approaches exist that aim at automating this task. However, predictions of state-of-the-art models for automatic cell-wise classification act as black-boxes and lack the explainability of human-created gating hierarchies. We propose a novel transformer-based approach that classifies cells in FCM data by mimicking the decision process of medical experts. Our network considers all events of a sample at once and predicts the corresponding polygons of the gating hierarchy, thus, producing a verifiable visualization in the same way a human operator does. The proposed model has been evaluated on three publicly available datasets for acute lymphoblastic leukemia (ALL). In experimental comparison it reaches state-of-the-art performance for automated blast cell identification while providing transparent results and explainable visualizations for human experts.

Monitoring of minimal residual disease (MRD) by flow cytometry (FCM) is a powerful prognostic tool for predicting outcomes in acute lymphoblastic leukemia (ALL). To apply FCM-MRD in large, collaborative trials, dedicated laboratory staff must be educated to concordantly high levels of expertise and their performance quality should be continuously monitored. We sought to install a unique and comprehensive training and quality control (QC) program involving a large number of reference laboratories within the international Berlin-Frankfurt-Münster (I-BFM) consortium, in order to complement the standardization of the methodology with an educational component and persistent quality control measures. Our QC and quality assurance (QA) program is based on four major cornerstones: (i) a twinning maturation program, (ii) obligatory participation in external QA programs (spiked sample send around, United Kingdom National External Quality Assessment Service (UK NEQAS)), (iii) regular participation in list-mode-data (LMD) file ring trials (FCM data file send arounds), and (iv) surveys of independent data derived from trial results. We demonstrate that the training of laboratories using experienced twinning partners, along with continuous educational feedback significantly improves the performance of laboratories in detecting and quantifying MRD in pediatric ALL patients. Overall, our extensive education and quality control program improved inter-laboratory concordance rates of FCM-MRD assessments and ultimately led to a very high conformity of risk estimates in independent patient cohorts.

Aim: Outcomes of children with high-risk (HR) relapsed acute lymphoblastic leukaemia (ALL) (N = 393), recruited to ALLR3 and ALL-REZ BFM 2002 trials, were analysed. Minimal residual disease (MRD) was assessed after induction and at predetermined time points until haematopoietic stem cell transplantation (SCT). Methods: Genetic analyses included karyotype, copy-number alterations and mutation analyses. Ten-year survivals were analysed using Kaplan-Meier and Cox models for multivariable analyses. Results: Outcomes of patients were comparable in ALLR3 and ALL-REZ BFM 2002. The eventfree survival of B-cell precursor (BCP) and T-cell ALL (T-ALL) was 22.6% and 26.2% (P = 0.94), respectively, and the overall survival (OS) was 32.6% and 28.2% (P = 0.11), respectively. Induction failures (38%) were associated with deletions of NR3C1 (P = 0.002) and BTG1 (P = 0.03) in BCP-ALL. The disease-free survival (DFS) and OS in patients with good vs poor MRD responses were 57.4% vs 22.6% (P < 0.0001) and 57.8% vs 32.0% (P = 0.0004), respectively. For BCP- and T-ALL, the post-SCT DFS and OS were 42.1% and 56.8% (P = 0.26) and 51.6% and 55.4% (P = 0.67), respectively. The cumulative incidences of post-SCT relapse for BCP- and T-ALL were 36.9% and 17.8% (P = 0.012) and of death were 10.7% and 25.5% (P = 0.013), respectively. Determinants of outcomes after SCT were acute graft versus host disease, preSCT MRD (>= 10(-3)), HR cytogenetics and TP53 alterations in BCP-ALL. Conclusion: Improvements in outcomes for HR ALL relapses require novel compounds in induction therapy to improve remission rates and immune targeted therapy after induction to maintain remission after SCT. (C) 2021 The Author(s). Published by Elsevier Ltd.

Activating mutations in cytosolic 59-nucleotidase II (NT5C2) are considered to drive relapse formation in acute lymphoblastic leukemia (ALL) by conferring purine analog resistance. To examine the clinical effects of NT5C2 mutations in relapsed ALL, we analyzed NT5C2 in 455 relapsed B-cell precursor ALL patients treated within the ALL-REZ BFM 2002 relapse trial using sequencing and sensitive allele-specific real-time polymerase chain reaction. We detected 110 NT5C2 mutations in 75 (16.5%) of 455 B-cell precursor ALL relapses. Two-thirds of relapses harbored subclonal mutations and only one-third harbored clonal mutations. Event-free survival after relapse was inferior in patients with relapses with clonal and subclonal NT5C2 mutations compared with those without (19% and 25% vs 53%, P <.001). However, subclonal, but not clonal, NT5C2 mutations were associated with reduced event-free survival in multivariable analysis (hazard ratio, 1.89; 95% confidence interval, 1.28-2.69; P = .001) and with an increased rate of nonresponse to relapse treatment (subclonal 32%, clonal 12%, wild type 9%, P <.001). Nevertheless, 27 (82%) of 33 subclonal NT5C2 mutations became undetectable at the time of nonresponse or second relapse, and in 10 (71%) of 14 patients subclonal NT5C2 mutations were undetectable already after relapse induction treatment. These results show that subclonal NT5C2 mutations define relapses associated with high risk of treatment failure in patients and at the same time emphasize that their role in outcome is complex and goes beyond mutant NT5C2 acting as a targetable driver during relapse progression. Sensitive, prospective identification of NT5C2 mutations is warranted to improve the understanding and treatment of this aggressive ALL relapse subtype.

Background:  Minimal residual disease (MRD) supports a more accurate assessment when children and adolescents with late bone marrow relapses of B-cell precursor (BCP) acute lymphoblastic leukaemia (ALL) will benefit from allogeneic haematopoietic stem cell transplantation (alloHSCT). A more detailed dissection of MRD heterogeneity, dynamics and the specific genetic aberrations involved promises to further improve treatment stratification. Methods:  Patients treated according to the ALL-REZ BFM 2002 trial/registry protocol with late BCP ALL bone marrow relapses (n=413) were included. AlloHSCT was indicated for patients with MRD≥10-3 after induction treatment (poor response). Findings: Patients with good (MRD<10-3) or poor responses to induction reached excellent event-free survival (EFS) and overall survival (OS) probabilities. However, EFS probability was decreased in patients with a very poor response (MRD≥10-2) to induction, in whom MRD persisted (≥10-4) until directly before alloHSCT more frequently than in intermediate poor responders (MRD≥10-3to<10-2). Patients with MRD≥10-3 prior to alloHSCT had the poorest prognoses, and were defined as late nonresponders. Patients ≥25% leukaemic blasts after induction had a very poor probability for EFS and OS. Multivariable Cox regression identified only TP53 alterations, in any MRD-based response subgroup, as independent prognostic marker. Interpretation: The excellent survival in patients with late relapsed BCP-ALL was achieved because of post-induction MRD-based treatment stratification of patients for alloHSCT. Two nonresponder cut-offs were identified that predict patients after induction who will not benefit from conventional treatment intensification and after consolidation who will not benefit from alloHSCT. They should be shunted toward controlled trials for novel therapies.   Funding Statement: German Childhood Cancer Foundation, German Jose Carreras Foundation. Declaration of Interests: The authors declare no conflict of interest. Ethics Approval Statement: The treatment protocols, including MRD assessments at the defined time points, were approved by local medical research ethics committees, and informed consent obtained from patients and/or guardians.

The emerging demand for personalized cancer diagnostics is in conflict with an increasing challenge for assay standardization by regulatory authorities. Minimal residual disease (MRD) monitoring of acute lymphoblastic leukemia (ALL) patients is such an assay [1]. Here, patient-individual primer designs and PCR optimizations make assay standardization challenging. The IRMA approach combines three technological advances to provide a solution for standardized and patient-individual PCR assays: (1) Software assisted assay design (2) Modular assay technology (3) Microfluidic standardization.

Accurate modeling of intratumor heterogeneity presents a bottleneck against drug testing. Flexibility in a preclinical platform is also desirable to support assessment of different endpoints. We established the model system, OHC‐NB1, from a bone marrow metastasis from a patient diagnosed with MYCN‐amplified neuroblastoma and performed whole‐exome sequencing on the source metastasis and the different models and passages during model development (monolayer cell line, 3D spheroid culture and subcutaneous xenograft tumors propagated in mice). OHC‐NB1 harbors a MYCN amplification in double minutes, 1p deletion, 17q gain and diploid karyotype, which persisted in all models. A total of 80–540 single‐nucleotide variants (SNVs) was detected in each sample, and comparisons between the source metastasis and models identified 34 of 80 somatic SNVs to be propagated in the models. Clonal reconstruction using the combined copy number and SNV data revealed marked clonal heterogeneity in the originating metastasis, with four clones being reflected in the model systems. The set of OHC‐NB1 models represents 43% of somatic SNVs and 23% of the cellularity in the originating metastasis with varying clonal compositions, indicating that heterogeneity is partially preserved in our model system.

Minimal residual disease (MRD) as measured by multiparameter flow cytometry (FCM) is an independent and strong prognostic factor in B‐cell acute lymphoblastic leukemia (B‐ALL). However, reliable flow cytometric detection of MRD strongly depends on operator skills and expert knowledge. Hence, an objective, automated tool for reliable FCM‐MRD quantification, able to overcome the technical diversity and analytical subjectivity, would be most helpful. We developed a supervised machine learning approach using a combination of multiple Gaussian Mixture Models (GMM) as a parametric density model. The approach was used for finding the weights of a linear combination of multiple GMMs to represent new, “unseen” samples by an interpolation of stored samples. The experimental data set contained FCM‐MRD data of 337 bone marrow samples collected at day 15 of induction therapy in three different laboratories from pediatric patients with B‐ALL for which accurate, expert‐set gates existed. We compared MRD quantification by our proposed GMM approach to operator assessments, its performance on data from different laboratories, as well as to other state‐of‐the‐art automated read‐out methods. Our proposed GMM‐combination approach proved superior over support vector machines, deep neural networks, and a single GMM approach in terms of precision and average F 1 ‐scores. A high correlation of expert operator‐based and automated MRD assessment was achieved with reliable automated MRD quantification ( F 1 ‐scores >0.5 in more than 95% of samples) in the clinically relevant range. Although best performance was found, if test and training samples were from the same system (i.e., flow cytometer and staining panel; lowest median F 1 ‐score 0.92), cross‐system performance remained high with a median F 1 ‐score above 0.85 in all settings. In conclusion, our proposed automated approach could potentially be used to assess FCM‐MRD in B‐ALL in an objective and standardized manner across different laboratories. © 2019 International Society for Advancement of Cytometry

PURPOSE:Minimal residual disease (MRD) helps to accurately assess when children with late bone marrow relapses of B-cell precursor (BCP) acute lymphoblastic leukemia (ALL) will benefit from allogeneic hematopoietic stem-cell transplantation (allo-HSCT). More detailed dissection of MRD response heterogeneity and the specific genetic aberrations could improve current practice.PATIENTS AND METHODS:MRD was assessed after induction treatment and at different times during relapse treatment until allo-HSCT (indicated in poor responders to induction; MRD ≥ 10-3) for patients being treated for late BCP-ALL bone marrow relapses (n = 413; median follow-up, 9.4 years) in the ALL-REZ BFM 2002 trial/registry (ClinicalTrials.gov identifier: NCT00114348).RESULTS:Patients with both good (MRD < 10-3) and poor responses to induction treatment reached excellent event-free survival (EFS; 72% v 65%) and overall survival (OS; 82% v 74%). Patients with MRD of 10-2 or greater after induction had reduced EFS (56%), and their MRD persisted until allo-HSCT more frequently than it did in patients with MRD of 10-3 or greater to less than 10-2 (P = .037). Patients with 25% or more leukemic blasts after induction (early nonresponders) had the poorest prognosis (EFS, 22%). Interestingly, patients with MRD of 10-3 or greater before allo-HSCT (late nonresponders) still had an EFS of 50% and OS of 63%, which in principle justifies allo-HSCT in these patients. From a panel of selected candidate genes, TP53 alterations (frequency, 8%) were the only genetic alteration with independent prognostic value in any MRD-based response subgroup.CONCLUSION:After induction treatment, MRD-based treatment stratification resulted in excellent survival in patients with late relapsed BCP-ALL. Prognosis could be further improved in very poor responders by intensifying treatment directly after induction. TP53 alterations can be defined as a novel genetic high-risk marker in all MRD response groups in late relapsed BCP-ALL. Here we identified early and late nonresponders to be considered as events in future trials.

Aneuploidy is common in paediatric B-cell precursor acute lymphoblastic leukaemia (ALL). Specific subgroups, such as high hyperdiploidy (>50 chromosomes or DNA Index ≥1·16) and hypodiploidy (<45 chromosomes), predict outcome of patients after primary treatment. Whether aneuploidy has a prognostic value for relapsed disease is yet to be determined. Using DNA index and centromere screening by multiplex ligation-dependent probe amplification, we investigated aneuploidy in 413 children treated for first relapse of B-cell precursor ALL according to the ALL-REZ BFM 2002 protocol. Ten-year event-free survival of patients with high hyperdiploid relapses approached 70%, whereas it was only 40% in low hyperdiploid relapses. Three patients with apparent hyperdiploid relapse had TP53 mutations. In these cases, array-based allelotyping revealed a hypodiploid origin with absence of the hypodiploid founder clone (masked hypodiploidy). Collectively, patients with evident or masked hypodiploid relapses showed an extremely low event-free survival rate of 9%. Importantly, the current relapse risk stratification did not identify cases with masked hypodiploidy as high-risk patients, due to their favourable clinical presentation. In multivariate analysis, hypodiploidy proved to be an independent prognostic factor. This finding supports stratification of relapses with hypodiploid origin into high-risk arms in future trials or allocation of patients to alternative treatment approaches.

Purpose Somatic deletions that affect the lymphoid transcription factor-coding gene IKZF1 have previously been reported as independently associated with a poor prognosis in pediatric B-cell precursor (BCP) acute lymphoblastic leukemia (ALL). We have now refined the prognostic strength of IKZF1 deletions by analyzing the effect of co-occurring deletions. Patients and Methods The analysis involved 991 patients with BCP ALL treated in the Associazione Italiana Ematologia ed Oncologia Pediatrica-Berlin-Frankfurt-Muenster (AIEOP-BFM) ALL 2000 trial with complete information for copy number alterations of IKZF1, PAX5, ETV6, RB1, BTG1, EBF1, CDKN2A, CDKN2B, Xp22.33/Yp11.31 (PAR1 region; CRLF2, CSF2RA, and IL3RA), and ERG; replication of findings involved 417 patients from the same trial. Results IKZF1 deletions that co-occurred with deletions in CDKN2A, CDKN2B, PAX5, or PAR1 in the absence of ERG deletion conferred the worst outcome and, consequently, were grouped as IKZF1plus. The IKZF1plus group comprised 6% of patients with BCP ALL, with a 5-year event-free survival of 53 ± 6% compared with 79 ± 5% in patients with IKZF1 deletion who did not fulfill the IKZF1plus definition and 87 ± 1% in patients who lacked an IKZF1 deletion ( P ≤ .001). Respective 5-year cumulative relapse incidence rates were 44 ± 6%, 11 ± 4%, and 10 ± 1% ( P ≤ .001). Results were confirmed in the replication cohort, and multivariable analyses demonstrated independence of IKZF1plus. The IKZF1plus prognostic effect differed dramatically in analyses stratified by minimal residual disease (MRD) levels after induction treatment: 5-year event-free survival for MRD standard-risk IKZF1plus patients was 94 ± 5% versus 40 ± 10% in MRD intermediate- and 30 ± 14% in high-risk IKZF1plus patients ( P ≤ .001). Corresponding 5-year cumulative incidence of relapse rates were 6 ± 6%, 60 ± 10%, and 60 ± 17% ( P ≤ .001). Conclusion IKZF1plus describes a new MRD-dependent very-poor prognostic profile in BCP ALL. Because current AIEOP-BFM treatment is largely ineffective for MRD-positive IKZF1plus patients, new experimental treatment approaches will be evaluated in our upcoming trial AIEOP-BFM ALL 2017.

Relapse of T-cell acute lymphoblastic leukemia (T-ALL) has a dismal prognosis, with only 20% of afflicted children surviving.[1][1] Children with relapsed T-ALL are commonly treated within high-risk arms of second-line treatment protocols that include mandatory hematopoietic stem cell

Intragenic PAX5 amplification defines a novel, relapse-prone subtype of B-cell precursor acute lymphoblastic leukemia with a poor outcome.

We propose a supervised learning approach to automatic quantification of cell populations in flow cytometric samples. One sample contains up to millions of measurement vectors with a dimensionality between 10 and 20. Normally, each measurement vector corresponds to a single cell in the biological sample. Identifying biologically meaningful cell populations is essentially a clustering problem, however, standard clustering methods are impractical, because size, shape and location of corresponding clusters may vary strongly between samples mainly due to phenotypic differences and inter-laboratory variations. In our holistic approach, we implicitly employ the structural information (such as relative locations and shape of sub-populations). A new input sample is reconstructed by a linear combination of artificial reference samples each represented by a Gaussian Mixture Model (GMM), in which for each Gaussian component the class label of the corresponding cluster of observations is known. The reference samples are calculated from a larger set of training samples by non-negative matrix factorization and can be regarded as the basis of a lower dimensional feature space, in which input samples are reconstructed. We show a method for calculating the feature space transformation based on minimization the L2 distance defined between two GMM. The feature space representation of the sample is then used to assign each observation to one of the specified sub-populations by a Bayes decision. We present classification results on a database of about 170 patients with Acute Lymphoblastic Leukemia (ALL), where high accuracy in the prediction of relatively small leukemic populations is crucial. The approach is not limited to our application. It can be employed wherever analysis of large, multi-dimensional, numerical data of a specific class of samples with related structure has to be performed.

In last years automated medical data analysis turned out to be one of the frontiers of Machine Learning. Medical operators are still reluctant to rely completely in automated solutions at diagnosis stage. However, Machine Learning researchers have focused their attention in this field, proposing valuable methods having often an outcome comparable to human evaluation. In this paper we give a brief overview on the role of Computer Vision and Machine Learning in solving medical problems in an automatic (supervised or unsupervised) fashion, we consider then a case study of Flow Cytometry data analysis for MRD assessment in Acute Lymphoblastic Leukemia. The clinical evaluation procedure of this type of data consists in a time taking manual labeling that can be performed only after an intensive training, however sometimes different experience may lead to different opinions. We are therefore proposing two different approaches: the first is generative semi-supervised Gaussian Mixture Model based approach, the latter is a discriminative semi-supervised Deep Learning based

Minimal Residual Disease (MRD) is a powerful risk-stratification marker in the treatment of primary and relapsed childhood Acute Lymphoblastic Leukemia. Flow Cytometry is a fast and sensitive method to detect MRD, however, the interpretation of these multi-parametric data demands intensive operator training and experience.