Abstract B046: Enhancing the Diagnosis and Monitoring of Chronic Lymphocytic Leukemia with Artificial Intelligence-Driven Peripheral Blood Smear Image Analysis
Bibliographic record
Abstract
Abstract Chronic lymphocytic leukemia (CLL), the most common leukemia in adults, presents with a broad morphological spectrum, particularly as the disease progresses. Peripheral blood (PB) smear review is essential for evaluating neoplastic cell morphology in CLL. Manual PB smear review, however, is labor-intensive, time-consuming, and subject to inter-observer variability. These limitations underscore the need for more objective and scalable diagnostic tools. Integrating artificial intelligence (AI) into CLL diagnostics may offer a promising solution for automating and standardizing PB smear evaluation. To develop the Convolutional Neural Network (CNN) models for CLL, we are expanding our current image database for MDA-LeukoLens system to include three more categories: (1) typical and atypical CLL cells, (2) prolymphocytes, and (3) large lymphoma cells. Given the variability in cell size across PB smears, we use the ratio of lymphocyte-to-RBC surface area, calculated as the square of the diameter ratio, for quantitative analysis. Approximately 10,000 CellaVision images per cell type are being collected for CNN model training. In addition to having one WBC, each CellaVision image also contains multiple RBCs. The image analysis model for this CLL project was used to create bounding boxes around all RBCs, and those with similar x- and y-axis diameters were selected to calculate the average RBC area per image. The area of each CNN-classified lymphocyte was then compared to the average RBC area in the same image by computing the ratio of the bounding box of the lymphocyte and the average bounding box area of the RBCs. RBCs in approximately 1,000 CellaVision images were labeled with bounding box to train the model for RBC cropping. An independent 1,001 CellaVision images containing atypical lymphoid cells from CLL patients were used to calculate the size ratio of lymphoid cells to RBCs. In the typical CLL cases (737 images), the size of atypical lymphoid cells was variable with the lymphocyte-to-RBC area ratio ranging from 0.5 to 4.49. Approximately 90% of lymphocytes were within the 1.0–2.5 range (1.0–1.49: 27.4%; 1.5–1.99: 42.9%; 2.0–2.49: 19.9%), with only 2.7% exceeding a ratio of 3.0. In the case of accelerated phase (67 images), the area ratio ranged from 2.0 to 7.99. 1.0–2.5: 6.0%; 2.5–4.0: 44.8%, 4.0–5.99: 40.3%, and 6.0–6.99: 7.5%. In the cases of Richter transformation (197 images), the area ratio ranged from 2.5 to 9.99, specifically, 1.0–2.5: 0%; 2.5–4.0: 10.2%; 4.0–5.99: 47.2%; and 6.0–9.99: 28.4%. In summary, our AI-assisted image analysis has revealed certain morphological variability among neoplastic B cells, even in classical CLL cases. A higher proportion of large lymphoid cells correlates with disease progression. This approach has demonstrated improved performance over existing automated imaging platforms and traditional reporting workflows. Overall, these findings support the utility of AI-assisted image analysis in reducing observer variability and providing a scalable framework for monitoring disease evolution and progression in CLL. Citation Format: J. Matthew. Liu, Yun Gong, Amaris Shi, Jeffrey Liu, Xiaoping Sun, Zhihong Hu. Enhancing the Diagnosis and Monitoring of Chronic Lymphocytic Leukemia with Artificial Intelligence-Driven Peripheral Blood Smear Image Analysis [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 B046.
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 | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.002 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 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".