Explainable AI identifies diagnostic cells of genetic AML subtypes

Explainable AI is deemed essential for clinical applications as it allows rationalizing model predictions, helping to build trust between clinicians and automated decision support tools. We developed an inherently explainable AI model for the classification of acute myeloid leukemia subtypes from blood smears and found that high-attention cells identified by the model coincide with those labeled as diagnostically relevant by human experts. Based on over 80,000 single white blood cell images from digitized blood smears of 129 patients diagnosed with one of four WHO-defined genetic AML subtypes and 60 healthy controls, we trained SCEMILA, a single-cell based explainable multiple instance learning algorithm. SCEMILA could perfectly discriminate between AML patients and healthy controls and detected the APL subtype with an F1 score of 0.86±0.05 (mean±s.d., 5-fold cross-validation). Analyzing a novel multi-attention module, we confirmed that our algorithm focused with high concordance on the same AML-specific cells as human experts do. Applied to classify single cells, it is able to highlight subtype specific cells and deconvolve the composition of a patient’s blood smear without the need of single-cell annotation of the training data. Our large AML genetic subtype dataset is publicly available, and an interactive online tool facilitates the exploration of data and predictions. SCEMILA enables a comparison of algorithmic and expert decision criteria and can present a detailed analysis of individual patient data, paving the way to deploy AI in the routine diagnostics for identifying hematopoietic neoplasms.Author summary: The analysis of blood and bone marrow smear microscopy by trained human experts remains an essential cornerstone of the diagnostic workup for severe blood diseases, like acute myeloid leukemia. While this step yields insight into a patient’s blood system composition, it is also tedious, time consuming and not standardized. Here, we present SCEMILA, an algorithm trained to distinguish blood smears from healthy stem cell donors and four different types of acute myeloid leukemia. Our algorithm is able to classify a patient’s blood sample based on roughly 400 single cell images, and can highlight cells most relevant to the algorithm. This allows us to cross-check the algorithm’s decision making with human expertise. We show that SCEMILA is able to identify relevant cells for acute myeloid leukemia, and therefore believe that it will contribute towards a future, where machine learning algorithms and human experts collaborate to form a synergy for high-performance blood cancer diagnosis.