Lung cancer remains the primary cause of cancer-related mortality worldwide, with carly detection posing substantial challenges due to the reliance on invasive diagnostic methods predominantly performed in laboratory settings. Despite advancements in microfluidic-chip applications for circulating tumor cell (CTC) analysis using liquid biopsy samples, most existing approaches rely on static image analysis, limiting their accuracy in dynamic fluid environments typical of physiological conditions. This study introduces a novel approach to dynamic magnetic bead analysis for CTC detection using an improved YOLOv8 model, improved with Squeeze-and-Excitation (SE) blocks, Global Attention Mechanism (GAM), and Soft Non-Maximum Suppression (Soft-NMS) post-processing. These architectural adjustments address spatial attention and feature recalibration challenges, optimizing model performance under dynamic fluidic conditions. The proposed model was rigorously tested on a microfluidic chip-based dynamic fluid platform, demonstrating superior detection accuracy over traditional static-based methodologies. Moreover, to enhance the adaptability of CTC detection for Point-of-Care Testing (POCT), this study also developed a portable device that integrates the dynamic fluid platform and an embedded computing unit to streamline real-time data processing and diagnosis. This device's lightweight design and embedded Al capabilities allow it to perform consistently in various settings, including low-resource environments. Comparative experimental results demonstrated that the proposed device achieved high performance, with counting accuracy exceeding 90% compared to ground truth data manually counted by trained attendants. The ratio of correctly identified counts to the total ground truth count was approximately 0.90:1, indicating a high level of agreement (90%) between the automated detection system and manual counting. The consistent accuracy, even under varying fluidic conditions, highlights" the model's robustness and its potential to reduce labor-intensive diagnostic procedures. These findings reinforce the practical viability of this approach for point-of-care testing (POCT), providing a promising solution for early-stage lung cancer diagnostics in decentralized healthcare settings, where accessible and non-invasive testing is critical for timely medical intervention. Furthermore, the integration of advanced Al techniques within a portable platform represents a significant advancement, making early cancer diagnostics more efficient, accessible, and scalable across diverse clinical environments.