Accurate cancer cell detection model for circulating tumor cell with telomerase reverse transcriptase

Telomerase reverse transcriptase (TERT) based circulating tumor cell (CTC) detection by flow cytometry is a promising innovation in the liquid biopsy. It provides high-throughput measurement of blood samples at single-cell level with much higher accuracy compared to the traditional CTC detection, promoting developments in many researches related to areas of cancer diagnosis/prognosis. However, to our best knowledge few methods are available to process data of TERT-based CTC detection except manual gating. This bottleneck is an important hindrance to the implementation of the TERT to wider pathological practices. To overcome this difficulty, we developed a cross fully connected neural network model, CFC-NN. In the experiment, our model outperformed the current state-of-the-art, attention-based model flowformer by a significant margin. Detailed interpretability analysis demonstrated that the attention mechanism in the flowformer could not recognize markers strongly related to the cancer cell. On the other hand, the CFC-NN model accurately identified all relative biomarkers. In summary, our work provides a highly accurate model for TERT based CTC detection.