Deep learning identifies Acute Promyelocytic Leukemia in bone marrow smears

Area of Science:

• Hematology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Acute promyelocytic leukemia (APL) is a life-threatening hematologic emergency with a high early death rate, particularly in vulnerable patients.
• Accurate and timely diagnosis is crucial for effective treatment with differentiation-inducing agents, leading to high cure rates.
• Current diagnostic methods, including cytomorphology and genetic confirmation of t(15;17), can be time-consuming and inaccessible in certain regions.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) platform for the automated diagnosis of APL using bone marrow smear images.
• To assess the DL platform's accuracy in segmenting bone marrow cells and classifying APL versus non-APL cases and healthy donors.

Main Methods:

• A multi-stage deep learning platform was developed to automatically read, segment, and classify bone marrow smear images.
• The study retrospectively analyzed 51 APL patients, 1048 non-APL acute myeloid leukemia (AML) patients, and 236 healthy bone marrow donors.
• The DL platform's performance was evaluated using image data alone, without genetic or cytogenetic information.

Main Results:

• The DL platform achieved high accuracy in segmenting bone marrow cells, with mean average precision and recall of 0.97.
• The platform demonstrated strong performance in APL detection, distinguishing APL from non-APL AML (AUC 0.8575) and healthy donors (AUC 0.9585).
• The DL model successfully inferred correct predictions from a limited dataset of 51 APL patients.

Conclusions:

• Deep learning is feasible for detecting APL based on distinct morphologies associated with the t(15;17) aberration.
• The DL platform's ability to abstract information from limited medical data highlights its potential for diagnosing rare cancer entities.
• This AI-driven approach can assist in APL diagnosis in resource-limited settings and flag suspected cases for expert review.