Atrial arrhythmias following CAR-chimeric antigen receptor T-cell therapy: Incidence, risk factors and biomarker profile.

Recent reports have raised concerns about the association of chimeric antigen receptor T cell (CAR-T) with non-negligible cardiotoxicity, particularly atrial arrhythmias. First, we conducted a pharmacovigilance study to assess the reporting of atrial arrhythmias following CD19-directed CAR-T. Subsequently, to determine the incidence, risk factors and outcomes of atrial arrhythmias post-CAR-T, we compiled a retrospective single-centre cohort of non-Hodgkin lymphoma patients. Only commercial CAR-T products were considered. Atrial arrhythmias were nearly fourfold more likely to be reported after CAR-T therapy compared to all other cancer patients in the FAERS (adjusted ROR = 3.76 [95% CI 2.67-5.29]). Of the 236 patients in our institutional cohort, 23 (10%) developed atrial arrhythmias post-CAR-T, including 12 de novo arrhythmias, with most (83%) requiring medical intervention. Atrial arrhythmias frequently co-occurred with cytokine release syndrome and were associated with higher post-CAR-T infusion peak levels of IL-10, TNF-alpha and LDH, and lower trough levels of fibrinogen. In a multivariable analysis, risk factors for atrial arrhythmia were history of atrial arrhythmia (OR = 6.80 [2.39-19.6]) and using CAR-T product with a CD28-costimulatory domain (OR = 5.17 [1.72-18.6]). Atrial arrhythmias following CD19-CAR-T therapy are prevalent and associated with elevated inflammatory biomarkers, a history of atrial arrhythmia and the use of a CAR-T product with a CD28 costimulatory domain.

Emapalumab as salvage therapy for adults with malignancy-associated hemophagocytic lymphohistiocytosis.

Not available.

Axicabtagene ciloleucel versus standard of care in second-line large B-cell lymphoma: outcomes by metabolic tumor volume.

Metabolic tumor volume (MTV) assessed using 2-deoxy-2-[18F]fluoro-D-glucose positron emission tomography, a measure of tumor burden, is a promising prognostic indicator in large B-cell lymphoma (LBCL). This exploratory analysis evaluated relationships between baseline MTV (categorized as low [≤median] vs high [>median]) and clinical outcomes in the phase 3 ZUMA-7 study (NCT03391466). Patients with LBCL relapsed within 12 months of or refractory to first-line chemoimmunotherapy were randomized 1:1 to axicabtagene ciloleucel (axi-cel; autologous anti-CD19 chimeric antigen receptor [CAR] T-cell therapy) or standard care (2-3 cycles of chemoimmunotherapy followed by high-dose chemotherapy with autologous stem-cell transplantation in patients who had a response). All P values are descriptive. Within high and low MTV subgroups, event-free survival (EFS) and progression-free survival (PFS) were superior with axi-cel vs standard care (all HR ≤0.523; P<.01). EFS in patients with high MTV (vs low MTV) was numerically shorter with axi-cel (HR, 1.448; P=.06) and was significantly shorter with standard care (HR, 1.486; P=.02). PFS was shorter in patients with high MTV vs low MTV in both the axi-cel (HR,1.660; P=.02) and standard-care (HR, 1.635; P=.02) arms, and median MTV was lower in patients in ongoing response at data cutoff vs others (both P≤.01). Median MTV was higher in axi-cel-treated patients who experienced grade ≥3 neurologic events or cytokine release syndrome (CRS) than in patients with grade 1/2 or no neurologic events or CRS, respectively (both P≤.03). Baseline MTV ≤median was associated with better clinical outcomes in patients receiving axi-cel or standard care for second-line LBCL.

ASCT2-Targeting Antibody-Drug Conjugate MEDI7247 in Adult Patients with Relapsed/Refractory Hematological Malignancies: A First-in-Human, Phase 1 Study.

BACKGROUND: MEDI7247 is a first-in-class antibody-drug conjugate (ADC) consisting of an anti-sodium-dependent alanine-serine-cysteine transporter 2 antibody-conjugated to a pyrrolobenzodiazepine dimer. OBJECTIVE: This first-in-human phase 1 trial evaluated MEDI7247 in patients with hematological malignancies. PATIENTS AND METHODS: Adults with acute myeloid leukemia (AML), multiple myeloma (MM), or diffuse large B-cell lymphoma (DLBCL) relapsed or refractory (R/R) to standard therapies, or for whom no standard therapy exists, were eligible. Primary endpoints were safety and determination of the maximum tolerated dose (MTD). Secondary endpoints included assessments of antitumor activity, pharmacokinetics (PK), and immunogenicity. RESULTS: As of 26 March 2020, 67 patients were treated (AML: n = 27; MM: n = 18; DLBCL: n = 22). The most common MEDI7247-related adverse events (AEs) were thrombocytopenia (41.8%), neutropenia (35.8%), and anemia (28.4%). The most common treatment-related grade 3/4 AEs were thrombocytopenia (38.8%), neutropenia (34.3%), and anemia (22.4%). Anticancer activity (number of responders/total patients evaluated) was observed in 11/67 (16.4%) patients. No correlation was observed between ASCT2 expression and clinical response. Between-patient variability of systemic exposure of MEDI7247 ADC and total antibody were high (AUCinf geometric CV%: 62.3-134.2, and 74.8-126.1, respectively). SG3199 (PBD dimer) plasma concentrations were below the limit of quantification for all patients after Study Day 8. Anti-drug antibody (ADA) prevalence was 7.7%, ADA incidence was 1.9%, and persistent-positive ADA was 5.8%. CONCLUSIONS: Thrombocytopenia and neutropenia limited repeat dosing. Although limited clinical activity was detected, the dose-escalation phase was stopped early without establishing an MTD. The study was registered with ClinicalTrials.gov (NCT03106428).

B-cell non-Hodgkin lymphomas.

B-cell lymphomas occur with an incidence of 20 new cases per 100 000 people per year in high-income countries. They can affect any organ and are characterised by heterogeneous clinical presentations and courses, varying from asymptomatic, to indolent, to very aggressive cases. Since the topic of B-cell non-Hodgkin lymphomas was last reviewed in The Lancet in 2017, a deeper understanding of the biological background of this heterogeneous group of malignancies, the availability of new diagnostic methods, and the development and implementation of new targeted and immunotherapeutic approaches have improved our ability to treat patients. This Seminar provides an overview of the pathobiology, classification, and prognostication of B-cell non-Hodgkin lymphomas and summarises the current knowledge and standard of care regarding biology and clinical management of the most common subtypes of mature B-cell non-Hodgkin lymphomas. It also highlights new findings in deciphering the molecular background of disease development and the implementation of new therapeutic approaches, particularly those targeting the immune system.

Conventional and novel [18F]FDG PET/CT features as predictors of CAR-T cell therapy outcome in large B-cell lymphoma.

Relapse and toxicity limit the effectiveness of chimeric antigen receptor T-cell (CAR-T) therapy for large B-cell lymphoma (LBCL), yet biomarkers that predict outcomes and toxicity are lacking. We examined radiomic features extracted from pre-CAR-T 18F-fluorodeoxyglucose positron emission tomography/computed tomography ([18F]FDG PET/CT) scans (n = 341) of 180 patients (121 male; median age, 66 years). Three conventional (maximum standardized uptake value [SUVmax], metabolic tumor volume [MTV], total lesion glycolysis [TLG]) and 116 novel radiomic features were assessed, along with inflammatory markers, toxicities, and outcomes. At both pre-apheresis and pre-infusion time points, conventional PET features of disease correlated with elevated inflammatory markers. At pre-infusion, MTV was associated with grade ≥ 2 cytokine release syndrome (odds ratio [OR] for 100 mL increase: 1.08 [95% confidence interval (CI), 1.01-1.20], P = 0.031), and SUVmax was associated with failure to achieve complete response (CR) (OR 1.72 [95% CI, 1.24-2.43], P < 0.001). Higher pre-apheresis and pre-infusion MTV values were associated with shorter progression-free survival (PFS) (HR for 10-unit increase: 1.11 [95% CI, 1.05-1.17], P < 0.001; 1.04 [95% CI, 1.02-1.07], P < 0.001) and shorter overall survival (HR for 100-unit increase: 1.14 [95% CI, 1.07-1.21], P < 0.001; 1.04 [95% CI, 1.02-1.06], P < 0.001). A combined MTV and LDH measure stratified patients into high and low PFS risk groups. Multiple pre-infusion novel radiomic features were associated with CR. These quantitative conventional [18F]FDG PET/CT features obtained before CAR-T cell infusion, which were correlated with inflammation markers, may provide prognostic biomarkers for CAR-T therapy efficacy and toxicity. The use of conventional and novel radiomic features may thus help identify high-risk patients for earlier interventions.

Cell-free DNA from nail clippings as source of normal control for genomic studies in hematologic malignancies.

Comprehensive genomic sequencing is becoming a critical component in the assessment of hematologic malignancies, with broad implications for patient management. In this context, unequivocally discriminating somatic from germline events is challenging but greatly facilitated by matched analysis of tumor:normal pairs. In contrast to solid tumors, conventional sources of normal control (peripheral blood, buccal swabs, saliva) could be highly involved by the neoplastic process, rendering them unsuitable. In this work we describe our real-world experience using cell free DNA (cfDNA) isolated from nail clippings as an alternate source of normal control, through the dedicated review of 2,610 tumor:nail pairs comprehensively sequenced by MSK-IMPACT-heme. Overall, we find nail cfDNA is a robust source of germline control for paired genomic studies. In a subset of patients, nail DNA may have tumor DNA contamination, reflecting unique attributes of the hematologic disease and transplant history. Contamination is generally low level, but significantly more common among patients with myeloid neoplasms (20.5%; 304/1482) compared to lymphoid diseases (5.4%; 61/1128) and particularly enriched in myeloproliferative neoplasms with marked myelofibrosis. When identified in patients with lymphoid and plasma-cell neoplasms, mutations commonly reflected a myeloid profile and correlated with a concurrent/evolving clonal myeloid neoplasm. For nails collected after allogeneic stem-cell transplantation, donor DNA was identified in 22% (11/50). In this cohort, an association with recent history of graft-vs-host disease was identified. These findings should be considered as a potential limitation for the use of nail as normal control but could also provide important diagnostic information regarding the disease process.

Matching-Adjusted Indirect Comparison of Brexucabtagene Autoleucel (ZUMA-2) and Pirtobrutinib (BRUIN) in Patients with Relapsed/Refractory Mantle Cell Lymphoma Previously Treated with a Covalent Bruton Tyrosine Kinase Inhibitor.

INTRODUCTION: Patients with relapsed/refractory (R/R) mantle cell lymphoma (MCL) often require multiple lines of treatment and have a poor prognosis, particularly after failing covalent Bruton tyrosine kinase inhibitor (cBTKi) therapy. Newer treatments such as brexucabtagene autoleucel (brexu-cel, chimeric antigen receptor T cell therapy) and pirtobrutinib (non-covalent BTKi) show promise in improving outcomes. METHODS: Without direct comparative evidence, an unanchored matching-adjusted indirect comparison was conducted to estimate the relative treatment effects of brexu-cel and pirtobrutinib for post-cBTKi R/R MCL. Using logistic propensity score models, individual patient-level data from ZUMA-2 brexu-cel-infused population (N = 68) were weighted to match pre-specified clinically relevant prognostic factors based on study-level data from the BRUIN cBTKi pre-treated cohort (N = 90). The base-case model incorporated the five most pertinent factors reported in ≥ 50% of both trial populations: morphology, MCL International Prognostic Index, number of prior lines of therapy, disease stage, and prior autologous stem cell transplant. A sensitivity analysis additionally incorporated TP53 mutation and Ki-67 proliferation. Relative treatment effects were expressed as odds ratios (ORs) or hazard ratios (HRs) with 95% confidence intervals (CIs). RESULTS: In the base-case model, brexu-cel was associated with higher rates of objective response (OR 10.39 [95% CI 2.81-38.46]) and complete response (OR 10.11 [95% CI 4.26-24.00]), and improved progression-free survival (HR 0.44 [95% CI 0.25-0.75]), compared to pirtobrutinib. Overall survival and duration of response favored brexu-cel over pirtobrutinib but the differences crossed the bounds for statistical significance. Findings were consistent across the adjusted and unadjusted analyses. CONCLUSIONS: Findings suggest that brexu-cel may offer clinically and statistically significant benefits regarding objective response, complete response, and progression-free survival compared to pirtobrutinib among patients with R/R MCL after prior cBTKi therapy. Given the short follow-up and high degree of censoring in BRUIN, an analysis incorporating updated BRUIN data may provide more definitive overall survival results.

Validation of LymphGen classification on a 400-gene clinical next-generation sequencing panel in diffuse large B-cell lymphoma: real-world experience from a cancer center.

Not available.

Matched control analysis suggests that R-CHOP followed by (R)-ICE may improve outcome in non-GCB DLBCL compared with R-CHOP.

ABSTRACT: Rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP) is considered the standard-of-care for patients with advanced-stage diffuse large B-cell lymphoma (DLBCL), despite findings that patients with nongerminal center B-cell like (non-GCB) have significantly worse outcome with this regimen. We evaluated the prognostic significance of baseline risk factors, including cell of origin (COO) classified by the Hans algorithm, within an alternative chemoimmunotherapy program. At Memorial Sloan Kettering Cancer Center (MSK), 151 patients with DLBCL received sequential R-CHOP induction and (R)-ICE (rituximab, ifosfamide, carboplatin, and etoposide) consolidation. Outcome analysis based on COO was validated with a propensity score-matched cohort treated with R-CHOP from the Mayo Clinic component of the Molecular Epidemiology Resource (MER). Among the patients with GCB (n = 69) and non-GCB (n = 69) at MSK, event-free survival (EFS) of non-GCB was superior to that of GCB (hazard ratio [HR], 0.53; 95% confidence interval [CI], 0.29-0.98). Overall survival (OS) demonstrated an association in the same direction but was not statistically significant (HR, 0.68; 95% CI, 0.33-1.42). Propensity score-matched patients from MSK (n = 108) demonstrated a small attenuation in the HRs for EFS (HR, 0.57; 95% CI, 0.27-1.18) and OS (HR, 0.76; 95% CI, 0.33-1.79) and were no longer statistically significant. In contrast, the matched MER cohort (n = 108) demonstrated an EFS association (HR, 1.17; 95% CI, 0.70-1.95) and OS association (HR, 1.13; 95% CI, 0.64-2.00) in the opposite direction, but were also not statistically significant. R-CHOP induction and (R)-ICE consolidation may overcome the negative prognostic impact of the non-GCB phenotype, per the Hans algorithm, and can be preferentially selected for this population. This trial was registered at www.ClinicalTrials.gov as #NCT00039195 and #NCT00712582.

Indirect treatment comparison of brexucabtagene autoleucel (ZUMA-2) versus standard of care (SCHOLAR-2) in relapsed/refractory mantle cell lymphoma.

The SCHOLAR-2 retrospective study highlighted poor overall survival (OS) with standard of care (SOC) regimens among patients with relapsed/refractory (R/R) mantle cell lymphoma (MCL) who failed a covalent Bruton tyrosine kinase inhibitor (BTKi). In the ZUMA-2 single-arm trial, brexucabtagene autoleucel (brexu-cel; autologous anti-CD19 CAR T-cell therapy) demonstrated high rates of durable responses in patients with R/R MCL who had previous BTKi exposure. Here, we compared OS in ZUMA-2 and SCHOLAR-2 using three different methods which adjusted for imbalances in prognostic factors between populations: inverse probability weighting (IPW), regression adjustment (RA), and doubly robust (DR). Brexu-cel was associated with improved OS compared to SOC across all unadjusted and adjusted comparisons. Hazard ratios (95% confidence intervals) were 0.38 (0.23, 0.61) for IPW, 0.45 (0.28, 0.74) for RA, and 0.37 (0.23, 0.59) for DR. These results suggest a substantial survival benefit with brexu-cel versus SOC in patients with R/R MCL after BTKi exposure.

Tafasitamab for patients with relapsed or refractory diffuse large B-cell lymphoma: final 5-year efficacy and safety findings in the phase II L-MIND study.

Tafasitamab, an anti-CD19 immunotherapy, is used with lenalidomide for patients with autologous stem cell transplant-ineligible relapsed/refractory diffuse large B-cell lymphoma based on the results of the phase II L-MIND study (NCT02399085). We report the final 5-year analysis of this study. Eighty patients ≥18 years who had received one to three prior systemic therapies, and had Eastern Cooperative Oncology Group performance status 0-2 received up to 12 cycles of co-administered tafasitamab and lenalidomide, followed by tafasitamab monotherapy until disease progression or unacceptable toxicity. The primary endpoint was the best objective response rate. Secondary endpoints included duration of response, progression-free survival, overall survival, and safety. Exploratory analyses evaluated efficacy endpoints by prior lines of therapy. At data cutoff on November 14, 2022, the objective response rate was 57.5%, with a complete response rate of 41.3% (n=33), which was consistent with prior analyses. With a median follow-up of 44.0 months, the median duration of response was not reached. The median progression-free survival was 11.6 months (95% confidence interval [95% CI]: 5.7-45.7) with a median follow-up of 45.6 months. The median overall survival was 33.5 months (95% CI: 18.3-not reached) with a median follow-up of 65.6 months. Patients who had received one prior line of therapy (n=40) showed a higher objective response rate (67.5%; 52.5% complete responses) compared to patients who had received two or more prior lines of therapy (n=40; 47.5%; 30% complete responses), but the median duration of response was not reached in either subgroup. Other exploratory analyses revealed consistent long-term efficacy results across subgroups. Adverse events were consistent with those described in previous reports, were manageable, and their frequency decreased during tafasitamab monotherapy, with no new safety concerns. This final 5-year analysis of L-MIND demonstrates that the immunotherapy combination of tafasitamab and lenalidomide is well tolerated and has long-term clinical benefit with durable responses.

Immunochemotherapy plus lenalidomide for high-risk mantle cell lymphoma with measurable residual disease evaluation.

Chemoimmunotherapy followed by consolidative high-dose therapy with autologous stem cell rescue was a standard upfront treatment for fit patients with mantle cell lymphoma (MCL) in first remission; however, treatment paradigms are evolving in the era of novel therapies. Lenalidomide is an immunomodulatory agent with known efficacy in treating MCL. We conducted a single-center, investigator-initiated, phase II study of immunochemotherapy incorporating lenalidomide, without autologous stem cell transplant consolidation, enriching for patients with high-risk MCL (clinicaltrials gov. Identifier: NCT02633137). Patients received four cycles of lenalidomide-R-CHOP, two cycles of R-HiDAC, and six cycles of R-lenalidomide. The primary endpoint was rate of 3-year progression-free survival. We measured measurable residual disease (MRD) using a next-generation sequencing-based assay after each phase of treatment and at 6 months following end-oftreatment. We enrolled 49 patients of which 47 were response evaluable. By intent-to-treat, rates of overall and complete response were equivalent at 88% (43/49), one patient with stable disease, and two patients had disease progression during study; 3-year progression-free survival was 63% (primary endpoint not met) and differed by TP53 status (78% wild-type vs. 38% ALT; P=0.043). MRD status was prognostic and predicted long-term outcomes following R-HiDAC and at 6 months following end-of-treatment. In a high-dose therapy-sparing, intensive approach, we achieved favorable outcomes in TP53- wild-type MCL, including high-risk cases. We confirmed that sequential MRD assessment is a powerful prognostic tool in patients with MCL.

Diffuse large B-cell lymphoma involving osseous sites: utility of response assessment by PET/CT and good longterm outcomes.

Osseous involvement by diffuse large B-cell lymphoma (DLBCL-bone) is a heterogeneous disease. There is limited data regarding response assessment by positron emission tomography with fluorodeoxyglucose, which may demonstrate residual avidity despite a complete response. We analyzed clinical data of patients with newly diagnosed DLBCL and identified all cases with DLBCL-bone. End of treatment scans were reviewed by two independent experts classifying osseous lesions into Deauville (DV) ≤3; DV ≥4, or reactive uptake in the bone marrow (M), site of fracture (F) or surgery (S). We compared outcomes of DLBCL-bone to other extranodal sites (EN) matched on International Prognotic Index features and regimen. Of 1,860 patients with DLBCL (bone 16%; EN 45%; nodal 39%), 41% had localized disease and 59% advanced. Only 9% (n=27) of patients with initial bone involvement had residual fluorodeoxyglucose avidity at the osseous site. In half of these cases, the uptake was attributed to F/S/M, and of the remaining 13, only two were truly refractory (both with persistent disease at other sites). Overall survival and progression-free survival (PFS) were found to be similar for early- stage nodal DLBCL and DLBCL-bone, but inferior in EN-DLBCL. Advanced-stage disease involving the bone had a similar 5-year PFS to nodal disease and EN-DLBCL. After matching for International Prognotic Index and treatment regiments, PFS between bone and other EN sites was similar. Osseous involvement in DLBCL does not portend a worse prognosis. End of treatment DV ≥4 can be expected in 5-10% of cases, but in the absence of other signs of refractory disease, may be followed expectantly.

Predictive Value of Minimal Residual Disease for Efficacy of Rituximab Maintenance in Mantle Cell Lymphoma: Results From the European Mantle Cell Lymphoma Elderly Trial.

PURPOSE: The outcome of older patients with mantle cell lymphoma (MCL) has improved by the introduction of immunochemotherapy, followed by rituximab (R)-maintenance. Assessment of minimal residual disease (MRD) represents a promising tool for individualized treatment decisions and was a prospectively planned part of the European MCL Elderly trial. We investigated how MRD status influenced the efficacy of R-maintenance and how MRD can enable tailored consolidation strategies. PATIENTS AND METHODS: Previously untreated patients with MCL age 60 years or older have been randomly assigned to R versus interferon-alpha maintenance after response to rituximab, fludarabine, cyclophosphamide (R-FC) versus rituximab, cyclophosphamide, doxorubicin, vincristine, prednisone (R-CHOP). MRD monitoring was performed by real-time quantitative polymerase chain reaction (qPCR) following EuroMRD guidelines. RESULTS: A qPCR assay with a median sensitivity of 1 × 10-5 could be generated in 80% of 288 patients in an international, multicenter, multilaboratory setting. More extensive tumor dissemination facilitated the identification of a molecular marker. The efficacy of R-maintenance in clinical remission was confirmed for MRD-negative patients at the end of induction in terms of progression-free survival (PFS; hazard ratio [HR], 0.38 [95% CI, 0.21 to 0.63]) and overall survival (OS; HR, 0.37 [95% CI, 0.20 to 0.68]), particularly in R-CHOP-treated patients (PFS-HR, 0.23 [95% CI, 0.10 to 0.52]; OS-HR, 0.19 [95% CI, 0.07 to 0.52]). R-maintenance appeared less effective in MRD-positive patients (PFS-HR, 0.51 [95% CI, 0.26 to 1.02]) overall and after R-CHOP induction (PFS-HR, 0.59 [95% CI, 0.28 to 1.26]). R-FC achieved more frequent and faster MRD clearance compared with R-CHOP. MRD positivity in clinical remission after induction was associated with a short median time to clinical progression of approximately 1-1.7 years. CONCLUSION: The results confirm the strong efficacy of R-maintenance in patients who are MRD-negative after induction. Treatment de-escalation for MRD-negative patients is discouraged by our results. More effective consolidation strategies should be explored in MRD-positive patients to improve their long-term prognosis.

Normal B cells express ZAP70 in chronic lymphocytic leukemia: A link between autoimmunity and lymphoproliferation?

ZAP70 has a prognostic value in chronic lymphocytic leukemia (CLL), through altered B-cell receptor signaling, which is important in CLL pathogenesis. A good correlation between ZAP70 expression in CLL cells and the occurrence of autoimmune phenomena has been reported. Yet, the great majority of CLL-associated autoimmune cytopenia is due to polyclonal immunoglobulin (Ig) G synthesized by nonmalignant B cells, and this phenomenon is poorly understood. Here, we show, using flow cytometry, that a substantial percentage of CD5- nonmalignant B cells from CLL patients expresses ZAP70 compared with CD5- B cells from healthy subjects. This ZAP70 expression in normal B cells from CLL patients was also evidenced by the detection of ZAP70 mRNA at single-cell level with polyclonal Ig heavy- and light-chain gene transcripts. ZAP70+ normal B cells belong to various B-cell subsets and their presence in the naïve B-cell subset suggests that ZAP70 expression may occur during early B-cell development in CLL patients and potentially before malignant transformation. The presence of ZAP70+ normal B cells is associated with autoimmune cytopenia in CLL patients in our cohort of patients, and recombinant antibodies produced from these ZAP70+ nonmalignant B cells were frequently autoreactive including anti-platelet reactivity. These results provide a better understanding of the implication of ZAP70 in CLL leukemogenesis and the mechanisms of autoimmune complications of CLL.

Deep learning for [18F]fluorodeoxyglucose-PET-CT classification in patients with lymphoma: a dual-centre retrospective analysis.

BACKGROUND: The rising global cancer burden has led to an increasing demand for imaging tests such as [18F]fluorodeoxyglucose ([18F]FDG)-PET-CT. To aid imaging specialists in dealing with high scan volumes, we aimed to train a deep learning artificial intelligence algorithm to classify [18F]FDG-PET-CT scans of patients with lymphoma with or without hypermetabolic tumour sites. METHODS: In this retrospective analysis we collected 16 583 [18F]FDG-PET-CTs of 5072 patients with lymphoma who had undergone PET-CT before or after treatment at the Memorial Sloa Kettering Cancer Center, New York, NY, USA. Using maximum intensity projection (MIP), three dimensional (3D) PET, and 3D CT data, our ResNet34-based deep learning model (Lymphoma Artificial Reader System [LARS]) for [18F]FDG-PET-CT binary classification (Deauville 1-3 vs 4-5), was trained on 80% of the dataset, and tested on 20% of this dataset. For external testing, 1000 [18F]FDG-PET-CTs were obtained from a second centre (Medical University of Vienna, Vienna, Austria). Seven model variants were evaluated, including MIP-based LARS-avg (optimised for accuracy) and LARS-max (optimised for sensitivity), and 3D PET-CT-based LARS-ptct. Following expert curation, areas under the curve (AUCs), accuracies, sensitivities, and specificities were calculated. FINDINGS: In the internal test cohort (3325 PET-CTs, 1012 patients), LARS-avg achieved an AUC of 0·949 (95% CI 0·942-0·956), accuracy of 0·890 (0·879-0·901), sensitivity of 0·868 (0·851-0·885), and specificity of 0·913 (0·899-0·925); LARS-max achieved an AUC of 0·949 (0·942-0·956), accuracy of 0·868 (0·858-0·879), sensitivity of 0·909 (0·896-0·924), and specificity of 0·826 (0·808-0·843); and LARS-ptct achieved an AUC of 0·939 (0·930-0·948), accuracy of 0·875 (0·864-0·887), sensitivity of 0·836 (0·817-0·855), and specificity of 0·915 (0·901-0·927). In the external test cohort (1000 PET-CTs, 503 patients), LARS-avg achieved an AUC of 0·953 (0·938-0·966), accuracy of 0·907 (0·888-0·925), sensitivity of 0·874 (0·843-0·904), and specificity of 0·949 (0·921-0·960); LARS-max achieved an AUC of 0·952 (0·937-0·965), accuracy of 0·898 (0·878-0·916), sensitivity of 0·899 (0·871-0·926), and specificity of 0·897 (0·871-0·922); and LARS-ptct achieved an AUC of 0·932 (0·915-0·948), accuracy of 0·870 (0·850-0·891), sensitivity of 0·827 (0·793-0·863), and specificity of 0·913 (0·889-0·937). INTERPRETATION: Deep learning accurately distinguishes between [18F]FDG-PET-CT scans of lymphoma patients with and without hypermetabolic tumour sites. Deep learning might therefore be potentially useful to rule out the presence of metabolically active disease in such patients, or serve as a second reader or decision support tool. FUNDING: National Institutes of Health-National Cancer Institute Cancer Center Support Grant.