Summary 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.
Graphical abstract
Highlights • Device capabilities and diagnostic approaches differ greatly in lymphoma MFC panels • Single laboratories generate too little data to train an AI model with high accuracy • Transfer learning across panels increases classification performance significantly • Merging MFC data from multiple tubes per sample increases the model's transferability
The bigger picture 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.
In lymphoma diagnostics, artificial intelligence (AI) can save time and cost by improving the accuracy in disease subtyping with multi-parameter flow cytometry (MFC) data. So far, AI has been limited to the MFC protocol that was used to train the models. We present a framework to extend AI to multiple MFC protocols using transfer learning (TL). We demonstrate that TL in combination with MFC data merging achieves higher performance for smaller training sizes.