Recently, a joint research team from Fudan University and Shanghai Turtle Technology Co., Ltd. published a groundbreaking paper in the internationally renowned journal Sensors & Actuators: B. Chemical, titled “R³Net: A three-phase neural network for noise elimination and accurate quantification in chip-based digital PCR” (https://www.sciencedirect.com/science/article/abs/pii/S0925400525008998). This achievement marks a significant milestone in Turtle Technology’s development in the BioAI (biological artificial intelligence) field and showcases its commitment to its mission of “Leading Life Science and Molecular Diagnostics into Digital Era.”
Traditional chip-based digital PCR (cdPCR) systems face inherent challenges such as fluorescence noise, dark field interference, and light leakage during experimental operations and image acquisition. These issues can significantly compromise the precision and reliability of the results. Existing image processing methods often struggle to cope with such complex noise, which has become a bottleneck for further advancement of digital PCR technology.
Turtle Technology’s R³Net (Recognition-Restoration-Reading Net) revolutionizes this space with a novel three-phase neural network architecture. It employs a sequential process consisting of noise recognition, image restoration, and chip reading. The first step uses a U-Net-based model to accurately identify noise regions and generate noise masks. Next, a novel spiking residual network (S-SRNet) performs image restoration. Finally, a lightweight YOLO-mini network executes high-precision quantification. The unique temporal input mechanism enables the model to effectively distinguish between foreground noise and background regions, enhancing noise processing precision. Additionally, the integration of a spiking neural network (SNN) not only boosts computational efficiency but also strengthens temporal sensitivity, offering a new solution for image processing in noisy environments.
The experimental performance is outstanding. The research team conducted extensive tests using DNA samples from lung cancer, COVID-19, and influenza viruses. R³Net significantly outperformed conventional methods across key indicators: it achieved an accuracy of 88.47% in noisy image scenarios, with image clarity and similarity metrics reaching 41.38 and 99.72%, respectively. In terms of speed, the lightweight algorithm maintained a high accuracy of 98.27% while reducing system resource consumption by over 98%. It processes a single image in just 1.3 seconds—meeting the rapid demands of clinical diagnostics.
This technological breakthrough not only advances the academic frontier but also holds immense commercial and market value. As a leading provider of digital PCR and digital molecular diagnostics solutions in China, Turtle Technology offers a full range of products from the BioDigital Mini to the fully automated SCI Digital Pro. The successful integration of AI technology injects powerful new capabilities into this product ecosystem. For end-users, the R³Net-enhanced platforms promise more stable and reliable results even in challenging lab environments.
The successful application of R³Net signals the beginning of an AI-driven era in digital PCR and marks the first step in Turtle Technology’s BioAI roadmap. As a co-founder of the global life science AI community BioBuddy, Turtle Technology is actively building a diverse product portfolio including the BioSalesAI digital assistant solution and a dedicated training center for life science AI. Researchers and industry partners interested in collaborating are welcome to connect.