A Physician-in-the-Loop Approach by Means of Machine Learning for the Diagnosis of Lymphocytosis in the Clinical Laboratory.

CONTEXT.­: The goal of the lymphocytosis diagnosis approach is its classification into benign or neoplastic categories. Nevertheless, a nonnegligible percentage of laboratories fail in that classification. OBJECTIVE.­: To design and develop a machine learning model by using objective data from the DxH 800 analyzer, including cell population data, leukocyte and absolute lymphoid counts, hemoglobin concentration, and platelet counts, besides age and sex, with classification purposes for lymphocytosis diagnosis. DESIGN.­: A total of 1565 samples were included from 10 different lymphoid categories grouped into 4 diagnostic categories: normal controls (458), benign causes of lymphocytosis (567), neoplastic lymphocytosis (399), and spurious causes of lymphocytosis (141). The data set was distributed in a 60-20-20 scheme for training, testing, and validation stages. Six machine learning models were built and compared, and the selection of the final model was based on the minimum generalization error and 10-fold cross validation accuracy. RESULTS.­: The selected neural network classifier rendered a global 10-class classification validation accuracy corresponding to 89.9%, which, considering the aforementioned 4 diagnostic categories, presented a diagnostic impact accuracy corresponding to 95.8%. Finally, a prospective proof of concept was performed with 100 new cases with a global diagnostic accuracy corresponding to 91%. CONCLUSIONS.­: The proposed machine learning model was feasible, with a high benefit-cost ratio, as the results were obtained within the complete blood count with differential. Finally, the diagnostic impact with high accuracies in both model validation and proof of concept encourages exploration of the model for real-world application on a daily basis.

Abnormal characteristic "round bottom flask" shape volume-based scattergram as a trigger to suspect persistent polyclonal B-cell lymphocytosis.

BACKGROUND AND AIMS: The diagnosis of persistent polyclonal B-cell lymphocytosis (PPBL) is often challenging because of the lack of features and the overlap with the peripheral expression of splenic marginal zone lymphomas (SMZL). To obtain new clues for PPBL detection and diagnosis, all data provided by the DxH 800 analyzer (including scatter and cell population data (CPD)) was exploited and combined using a machine learning (ML) approach. MATERIALS AND METHODS: A total 211 samples from 101 normal controls and 110 patients (PPBL and SMZL) were assessed. Age, gender, full blood count, CPD, scatter, flags and CellaVision differentials were also considered. A ML model was built for true classification purposes. RESULTS: PPBL and SMZL shared increased absolute lymphoid counts, atypical lymphoid flag presence and CPD values (8 out of 14). A typical "round-bottom-flask" shape scattergram was described for the first time for PPBL which was also present in 51.4% of SMZL cases. The developed ML model render a global classification accuracy of 93.4%, allowing the detection of all pathological cases, with mean misclassification rates of 12% among PPBL and SMZL. CONCLUSION: Our ML model represents a new unbiased tool than can be widely applied in the laboratory as an aid for detection of PPBL.

Machine learning algorithms for the detection of spurious white blood cell differentials due to erythrocyte lysis resistance.

AIMS: Red blood cell (RBC) lysis resistance interferes with white blood cell (WBC) count and differential; still, its detection relies on the identification of an abnormal scattergram, and this is not clearly adverted by specific flags in the Beckman-Coulter DXH-800. The aims were to analyse precisely the effect of RBC lysis resistance interference in WBC counts, differentials and cell population data (CPD) and then to design, develop and implement a novel diagnostic machine learning (ML) model to optimise the detection of samples presenting this phenomenon. METHODS: WBC counts, differentials and CPD from 232 patients (anaemia or liver disease) were compared with 100 healthy controls (HC) using analysis of variance. The data were analysed after a corrective action, and the analyser differentials were also compared with the digital leucocyte differentials. The ML support vector machine (SVM) algorithm was trained with 70% of the samples (n=233) and the 30% remaining (n=99) were employed exclusively during the validation phase. RESULTS: We identified that impedance WBC was not affected by the RBC lysis resistance interference while the DXH-800 differentials overestimated lymphoid subpopulations (17.6%), sometimes even yielding spurious lymphocytosis, and the latter were corrected when sample dilution was performed. The ML-SVM algorithm allowed the classification of the pathological groups when compared with HC with validation accuracies corresponding to 97.98%, 100% and 88.78% for the global, anaemia and liver disease groups, respectively. CONCLUSIONS: The proposed algorithm has an impressive discriminatory potential and its application would be a valuable support system to detect spurious results due to RBC lysis resistance.