Knowledge transfer to enhance the performance of deep learning models for automated classification of B cell neoplasms.

Multi-parameter flow cytometry (MFC) is a cornerstone in clinical decision making for leukemia and lymphoma. MFC data analysis requires manual gating of cell populations, which is time-consuming, subjective, and often limited to a two-dimensional space. In recent years, deep learning models have been successfully used to analyze data in high-dimensional space and are highly accurate. However, AI models used for disease classification with MFC data are limited to the panel they were trained on. Thus, a key challenge in deploying AI into routine diagnostics is the robustness and adaptability of such models. This study demonstrates how transfer learning can be applied to boost the performance of models with smaller datasets acquired with different MFC panels. We trained models for four additional datasets by transferring the features learned from our base model. Our workflow increased the model's overall performance and, more prominently, improved the learning rate for small training sizes.

Influence of DM-sensitivity on immunogenicity of MHC class II restricted antigens.

BACKGROUND: Graft-versus-host-disease (GvHD) is a major problem in allogeneic stem cell transplantation. We previously described two types of endogenous human leukocyte antigen (HLA)-II restricted antigens depending on their behavior towards HLA-DM. While DM-resistant antigens are presented in the presence of HLA-DM, DM-sensitive antigens rely on the expression of HLA-DO-the natural inhibitor of HLA-DM. Since expression of HLA-DO is not upregulated by inflammatory cytokines, DM-sensitive antigens cannot be presented on non-hematopoietic tissues even under inflammatory conditions. Therefore, usage of CD4+ T cells directed against DM-sensitive antigens might allow induction of graft-versus-leukemia effect without GvHD. As DM-sensitivity is likely linked to low affinity peptides, it remains elusive whether DM-sensitive antigens are inferior in their immunogenicity. METHODS: We created an in vivo system using a DM-sensitive and a DM-resistant variant of the same antigen. First, we generated murine cell lines overexpressing either H2-M or H2-O (murine HLA-DM and HLA-DO) to assign the two model antigens ovalbumin (OVA) and DBY to their category. Further, we introduced mutations within the two T-cell epitopes and tested the effect on DM-sensitivity or DM-resistance. Furthermore, we vaccinated C57BL/6 mice with either variant of the epitope and measured expansion and reactivity of OVA-specific and DBY-specific CD4+ T cells. RESULTS: By testing T-cell recognition of OVA and DBY on a murine B-cell line overexpressing H2-M and H2-O, respectively, we showed that OVA leads to a stronger T-cell activation in the presence of H2-O demonstrating its DM-sensitivity. In contrast, the DBY epitope does not rely on H2-O for T-cell activation indicating DM-resistance. By introducing mutations within the T-cell epitopes we could generate one further DM-sensitive variant of OVA and two DM-resistant counterparts. Likewise, we designed DM-resistant and DM-sensitive variants of DBY. On vaccination of C57BL/6 mice with either epitope variant we measured comparable expansion and reactivity of OVA-specific and DBY-specific T-cells both in vivo and ex vivo. By generating T-cell lines and clones of healthy human donors we showed that DM-sensitive antigens are targeted by the natural T-cell repertoire. CONCLUSION: We successfully generated DM-sensitive and DM-resistant variants for two model antigens. Thereby, we demonstrated that DM-sensitive antigens are not inferior to their DM-resistant counterpart and are therefore interesting tools for immunotherapy after allogeneic stem cell transplantation.

Risk factors and mortality in invasive Rasamsonia spp. infection: Analysis of cases in the FungiScope® registry and from the literature.

BACKGROUND: The new Rasamsonia spp. complex can develop invasive infection in immunosuppression or chronic pulmonary disease. It has potential to be misidentified as other genera due to morphological similarities. Nowadays, there is a gap of knowledge on this fungi. OBJECTIVES: To provide knowledge base of risk factors and therapeutic decisions in invasive Rasamsonia spp. complex infection. PATIENTS/METHODS: Cases of invasive infection due to Rasamsonia spp. (formerly Geosmithia/Penicillium spp.) from FungiScope® registry and all reported cases from a literature were included. RESULTS: We identified 23 invasive infections due to Rasamsonia spp., six (26.1%) in the FungiScope® registry. Main risk factors were chronic granulomatous disease (n = 12, 52.2%), immunosuppressive treatment (n = 10, 43.5%), haematopoietic stem cell transplantation (n = 7, 30.4%), graft-versus-host disease and major surgery (n = 4, 17.4%, each). Predominantly affected organs were the lungs (n = 21, 91.3%), disease disseminated in seven cases (30.4%). Fungal misidentification occurred in 47.8% (n = 11), and sequencing was used in 69.6% of the patients (n = 16) to diagnose. Breakthrough infection occurred in 13 patients (56.5%). All patients received antifungal treatment, mostly posaconazole (n = 11), caspofungin (n = 10) or voriconazole (n = 9). Combination therapy was administered in 13 patients (56.5%). Susceptibility testing showed high minimum inhibitory concentrations for azoles and amphotericin B, but not for echinocandins. No preferable treatment influencing favourable outcome was identified. Overall mortality was 39% (n = 9). CONCLUSION: Rasamsonia spp. are emerging fungi causing life-threatening infections, especially in immunocompromised and critically ill patients. Mortality is high. Treatment is challenging and clinicians dealing with this patient population should become aware of this infection constituting a medical emergency.

A Comparison of Aspergillus and Mucorales PCR Testing of Different Bronchoalveolar Lavage Fluid Fractions from Patients with Suspected Invasive Pulmonary Fungal Disease.

In patients with hematological malignancies, bronchoalveolar lavage fluid (BALF) specimens are commonly used for the diagnosis of mold infections. However, it is not clear whether the cell pellet (P) or the supernatant fraction (S) of the BALF specimen is optimal for molecular diagnostic testing. Thus, 99 BALF specimens were collected from 96 hematology patients with or without allogeneic hematopoietic stem cell transplant. The cell pellets and supernatants were processed alone and in combination (S/P) for testing by two fungus-specific real-time PCR assays compliant with international recommendations. The results achieved with S/P were revealed to be superior in comparison to those achieved with S and P alone, with the use of each single fraction showing a reduced sensitivity for the detection of Aspergillus DNA (82% and 43% for S and P, respectively). In 57% of the samples, testing of the combination of S and P generated a lower quantification cycle value than testing of S or P alone. Molds would have been missed in 5 and 16 out of 28 samples if only S or P, respectively, was analyzed. No sample was positive by testing of S or P only. Similar results were obtained for the detection of Mucorales DNA in BALF specimens (reduced sensitivity of 67% and 50% for S and P, respectively). Study patients were categorized according to the current European Organization for the Research and Treatment of Cancer/Mycoses Study Group classification for invasive fungal disease (IFD), revealing that 35 patients had proven/probable IFD (36%), 47 patients had possible IFD (49%), and 14 patients had undetermined IFD (15%).

Measurements of immature platelets with haematology analysers are of limited value to separate immune thrombocytopenia from bone marrow failure.

Detection of immature platelets in the circulation may help to dissect thrombocytopenia due to platelet destruction from bone marrow failure (BMF). We prospectively tested the predictive value of immature platelets, measured as immature platelet fraction (IPF) on the XE-5000 (Sysmex, Kobe, Japan) or percentage of reticulated platelets (rPT) on the CD Sapphire (Abbott Diagnostics, Santa Clara, CA, USA) to separate immune thrombocytopenia (ITP) from BMF (leukaemia, myelodysplastic syndrome, aplastic anaemia). We analysed 58 samples of patients with BMF, 47 samples of patients with ITP and 97 controls. Median rPT (CD Sapphire) was increased to 9·0% in ITP and to 10·9% in BMF, compared to 1·9% in controls. Median IPF (XE-5000) was 16·2% in ITP, 10·2% in BMF and 2·5% in controls. We found an inverse correlation between high fractions of immature platelets and low platelet counts in thrombocytopenic samples regardless of the diagnosis. In conclusion, we observed a broad overlap of immature platelets between ITP and BMF, which may be caused by an accelerated release of immature platelets in any thrombocytopenic state and decreased production in many patients with ITP. Despite this, IPF (XE-5000) had some power to discriminate ITP from BMF, whereas rPT (CD Sapphire) was of no predictive value.

Measurement of immature platelets with Abbott CD-Sapphire and Sysmex XE-5000 in haematology and oncology patients.

BACKGROUND: Measurement of immature platelets was introduced into routine diagnostics by Sysmex as immature platelet fraction (IPF) some years ago and recently by Abbott as reticulated platelet fraction (rPT). Here, we compare both methods. METHODS: We evaluated the precision and agreement of these parameters between Sysmex XE-5000 and Abbott CD-Sapphire in three distinct thrombocytopaenic cohorts: 30 patients with beginning thrombocytopaenia and 64 patients with recovering platelets (PLT) after chemotherapy, 16 patients with immune thrombocytopaenia (ITP) or heparin-induced thrombocytopaenia type 2 (HIT) and 110 additional normal controls. Furthermore, we analysed, how IPF/rPT differed between these thrombocytopaenic cohorts and controls. RESULTS: Both analysers demonstrated acceptable overall precision (repeatability) of IPF/rPT with lower precision at low PLT counts. IPF/rPT artificially increased during storage of blood samples overnight. Inter-instrument comparison showed a moderate correlation (Pearson r²=0.38) and a systematic bias of 1.04 towards higher IPF-values with the XE-5000. IPF/rPT was highest in recovering thrombopoesis after chemotherapy and moderately increased in ITP/HIT. The normal range deduced from control samples was much narrower with CD-Sapphire (1.0%-3.8%, established here for the first time) in comparison to XE-5000 (0.8%-7.9%) leading to a smaller overlap of samples with increased PLT turnover and normal controls. CONCLUSIONS: IPF and rPT both give useful information on PLT turnover, although the two analysers only show a moderate inter-instrument correlation and have different reference ranges. A better separation of patient groups with high PLT turnover like ITP/HIT from normal controls is obtained by CD-Sapphire.

Comparison of automated differential blood cell counts from Abbott Sapphire, Siemens Advia 120, Beckman Coulter DxH 800, and Sysmex XE-2100 in normal and pathologic samples.

Reliable automated blood cell characterization and quantification remain challenging in pathologic samples, whereas slide reviews due to unnecessary flagging should be avoided. We compared 4 modern hematology analyzers-Abbott Sapphire, Siemens Advia 120, Sysmex XE-2100, and Beckman Coulter DxH 800-regarding complete blood cell count (CBC), leukocyte differential count, and flagging efficacy in a total of 202 samples from hematology patients and normal controls. Manual differential count was used as reference. The analyzers exhibited very good correlation for CBC parameters. Neutrophils and eosinophils also showed very good correlations, whereas lymphocytes and monocytes correlated fairly. The Advia 120 displayed notably lower measurements for both parameters, which is attributable to classification of some events as large unstained cells. Basophil counts were unreliable with all analyzers. Flagging for blasts and immature granulocytes showed moderate sensitivity and specificity. Operators must not rely on blast flagging alone to detect leukemic samples with any analyzer.