Technology and performance behind hema.to has been described in a number of scientific publications.
Nanditha Mallesh, Max Zhao, Lisa Meintker, Alexander Höllein, Franz Elsner, Hannes Lüling, Torsten Haferlach, Wolfgang Kern, Jörg Westermann, Peter Brossart, Stefan W. Krause, Peter M. Krawitz
Multi-parameter flow cytometry (MFC) is a critical tool in leukemia and lymphoma diagnostics. Advances in cytometry technology and diagnostic standardization efforts have led to an ever-increasing volume of data, presenting an opportunity to use artificial intelligence (AI) in diagnostics. However, the MFC protocol is prone to changes depending on the diagnostic workflow and the available cytometer. The changes to the MFC protocol limit the deployment of AI in routine diagnostics settings. We present a workflow that allows existing AI to adapt to multiple MFC protocols. We combine transfer learning (TL) with MFC data merging to increase the robustness of AI. Our results show that TL improves the performance of AI and allows models to achieve higher performance with less training data. This gain in performance for smaller training data allows for an already deployed AI to adapt to changes without the need for retraining a new model that requires more training data.
|
Cell Patterns |
https://www.cell.com/patterns/fulltext/S2666-3899(21)00206-3 |
Max Zhao, Nanditha Mallesh, Richard Schabath, PhD, Alexander Hoellein, MD, Claudia Haferlach, MD, Torsten Haferlach, MD, Franz Elsner, PhD, Hannes Lueling, PhD, Peter Krawitz, MD, Wolfgang Kern, MD
The wealth of information captured by multiparameter flow cytometry (MFC) can be analyzed by recent methods of computer vision when represented as a single image file. We therefore transformed MFC raw data into a multicolor 2D image by a self-organizing map (SOM) and classified this representation using a convolutional neural network (CNN). By this means, we built an artificial intelligence that is not only able to distinguish diseased from healthy samples, but that can also differentiate seven subtypes of mature B-cell neoplasm (B-NHL). We trained our model with 18,274 cases including chronic lymphocytic leukemia (CLL) and its precursor monoclonal B-cell lymphocytosis (MBL), marginal zone lymphoma (MZL), mantle cell lymphoma (MCL), prolymphocytic leukemia (PL), follicular lymphoma (FL), hairy cell leukemia (HCL), lymphoplasmacytic lymphoma (LPL) and achieved a weighted F1 score of 0.94 on a separate test set of 2,348 cases. Furthermore, we estimated the trustworthiness of a classification and could classify 70% of all cases with a confidence of 0.95 and higher. Our performance analyses indicate that particularly for rare subtypes further improvement can be expected when even more samples are available for training.
|
Cytometry: Part A |
Wolfgang Kern, MD, Franz Elsner, PhD, Max Zhao, Nanditha Mallesh, Richard Schabath, PhD, Claudia Haferlach, MD, Peter Krawitz, MD, Hannes Lueling, PhD, Torsten Haferlach, MD
Flow cytometry is an essential method in routine diagnostics for hematologic malignancies and is highly relevant in mature B-cell neoplasms. While flow cytometric procedures of sample preparation and measurement are subject to strict quality control mechanisms, analysis and interpretation of data are still vastly relying on expert knowledge applied to individual patient samples. To lower the dependency on expert knowledge and thus the inter-observer variability, we developed an automated mechanisms based on artificial intelligence.
|
Blood (2019) 134 (Supplement_1): 886. |
Nanditha Mallesh, Max Zhao, Franz Elsner, PhD, Hannes Lueling, PhD, Richard Schabath, PhD, Claudia Haferlach, MD, Torsten Haferlach, MD, Peter Krawitz, MD, Wolfgang Kern, MD
In multicolor flow cytometry (MFC), upgrading to a device that supports more fluorochromes per measurement is a common process in diagnostic laboratories. Usually this involves a period of applying both protocols during a transition period, the old one validating the new one. However, only very few of the rare diagnoses will be assessed by both protocols. In view of machine learning algorithms, typically this is too few data to train a new model to high accuracy. We therefore developed a new approach to successfully apply existing models under new setups.
|
Blood (2019) 134 (Supplement_1): 2094. |