Improving Circulating Tumor Cell Detection Using Image Synthesis and Transformer Models in Cancer Diagnostics

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Oncology
• Medical Imaging Analysis

Background:

• Cancer is a leading cause of death, with advanced stages lacking effective treatments.
• Early cancer diagnosis is critical for reducing mortality rates.
• Circulating tumor cells (CTCs) show promise for early detection but are difficult to automatically identify due to heterogeneity and scarcity.

Purpose of the Study:

• To develop an automated method for detecting circulating tumor cells (CTCs) for early cancer diagnosis.
• To address the challenges of CTC detection, including size/shape heterogeneity and low abundance.
• To improve the robustness and generalizability of CTC detection models.

Main Methods:

• Utilized the Segment Anything Model (SAM) and a copy-paste strategy for synthetic data generation.
• Developed a Swin Transformer-based detection network with specialized adapter modules for scale and shape.
• Implemented an improved loss function incorporating a regularization term for data distribution consistency.

Main Results:

• Achieved high performance metrics: accuracy (0.9960), recall (0.9961), precision (0.9804), specificity (0.9975), and mAP (0.9400 at IoU 0.5).
• Demonstrated robustness and generalizability on a mixed dataset of public and local data.
• Outperformed state-of-the-art models including ADCTC, DiffusionDet, CO-DETR, and DDQ.

Conclusions:

• The proposed framework offers a significant advancement in automated CTC detection.
• This technology can serve as a vital tool for early cancer diagnosis, treatment planning, and prognostic assessment.
• The approach has the potential to enhance patient outcomes and overall human well-being.