Abstract B031: Artificial Intelligence-Enhanced Image Analysis of Peripheral Blood Smears Supports the Diagnosis and Monitoring of Acute Promyelocytic Leukemia
Bibliographic record
Abstract
Abstract Acute promyelocytic leukemia (APL) is a high-risk leukemia requiring prompt diagnosis and treatment to reduce early mortality. Peripheral blood (PB) smears often provide excellent monolayer morphology, offering diagnostic information comparable to bone marrow aspirate smears. CellaVision, an automated imaging system widely used in hematology laboratories, enables automated classification of blood cells and generates high-resolution images to support accurate morphological assessment. To enhance early detection and precise monitoring of APL from PB smears, we developed a “MDA-LeukoLens” system, which combines two convolutional neural network (CNN) models (Model A and Model B) with a rule-based algorithm. Model A crops cells to minimize background noise, and Model B classifies the cropped CellaVision images. Both models were trained with the following configurations: 20 epochs, batch size of 8. After training, the models were evaluated using the following tools: Confusion Matrix, Precision-Recall Curve, Area Under the Curve (AUC), confidence interval (CI), and F1 score. For Model A, 11,000 CellaVision images were annotated, with bounding boxes drawn to identify cells for detection. Model A achieved an accuracy of over 99% in detecting and cropping white blood cells. For the development of model B, the database consists of over 113,236 CellaVision images representing 10 distinct cell types: abnormal promyelocytes, 10,196; basophils, 10,247; eosinophils, 10,201; erythroblasts/nucleated red blood cells (nRBCs), 10,288; left-shifted granulocytes, 11,000; lymphocytes, 10,104; monocytes, 10,012; myeloblasts including monoblasts, 10,656; neutrophils, 11,472; and plasma cells, 8,863. Smudge cells (n=10,197) were also included. This dataset was divided into training (83,679 images), validation (16,735 images), and testing (11,159 images) subsets. Model B achieved a validation classification accuracy of 96.2%. The rule-based algorithm incorporating four parameters (new diagnosis, characteristic Auer rods, strong myeloperoxidase expression, and elevated D-dimer level) further differentiate abnormal promyelocytes (APL cells) from their morphologic mimickers. The MDA-LeukoLens system was evaluated on 22,359 cells from 216 internal (MDACC) cases (148 APL and 68 non-APL) and 14,375 cells from 101 external cases (35 APL and 66 non-APL) from four other institutions. For APL cell classification, the sensitivity and accuracy were 0.965 and 0.991, respectively, for internal cases, and 0.965 and 0.981, respectively, for external cases. Notably, the system achieved 100% diagnostic accuracy for APL in both internal and external cases. Overall, CellaVision images provide diagnostic morphological evidence, which the MDA-LeukoLens system utilizes to identify APL cells with a classification performance comparable to the experienced hematopathologists. Through accurate classification, MDA-LeukoLens support the diagnosis and monitoring of APL by analyzing cell images from PB smears, thereby reducing diagnostic delays and facilitating early initiation of ATRA therapy. Citation Format: Xiaoping Sun, Amaris Shi, Jinchun Matthew. Liu, Jeffrey Liu, Yun Gong, Zhihong Hu. Artificial Intelligence-Enhanced Image Analysis of Peripheral Blood Smears Supports the Diagnosis and Monitoring of Acute Promyelocytic Leukemia [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr B031.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.003 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".