DOI 10.3390/diagnostics16081244
A Deep Learning Model for IMMP-Based Residual Disease Monitoring in AML with Monocytic Differentiation.
Аннотация
Background
Acute myeloid leukemia (AML) with monocytic differentiation poses significant clinical challenges, including high relapse rates and chemotherapy resistance. Current morphological assessment is limited by inter-observer variability, low sensitivity, and inefficiency, especially for detecting low-level residual disease. This creates an urgent need for automated, objective tools to improve diagnostic consistency and monitoring. Artificial intelligence, particularly deep learning, offers potential for extracting high-dimensional cytomorphological features to address these gaps.
Methods
A retrospective cohort of 184 bone marrow smear slides from patients with monocytic leukemia was used. The core biomarker was the immature monocyte percentage (IMMP), defined as monoblasts plus promonocytes among nucleated cells, with a 2.0% clinical cutoff. An EfficientNet-based convolutional neural network was developed via transfer learning and trained to classify four cell types: monoblasts, promonocytes, monocytes, and other cells.
Results
The model achieved robust cell-level classification, with F1 scores of 0.82 for monoblasts and 0.34 for promonocytes. At the slide level, using an optimized IMMP threshold of 0.045, it accurately assessed persistent leukemic cell burden with 78.9% Accuracy, 81.1% Recall, and 76.9% Specificity. Model-predicted IMMP values showed strong correlation with expert-derived values (Pearson r = 0.827), demonstrating reliable quantitative agreement.
Conclusions
This deep learning model provides an automated, objective tool for quantifying immature monocytes, addressing key limitations in morphological assessment of monocytic AML. The IMMP metric shows promise for monitoring treatment response, predicting relapse, and potentially identifying patients at risk of venetoclax-based therapy resistance. While promising, prospective multicenter validation is needed to translate these findings into routine clinical practice.
Подготовка и перевод публикации
Обработка статьи · этапы, ошибки и управление
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Тип публикации указан по данным источника. Уровень доказательности автоматически не присваивается.