Fluorescence imaging and recognition are core technologies in targeted medicine, pathological surgery, and biomedicine. However, current imaging and recognition systems are separate, requiring repeated data transfers for imaging and recognition that lead to delays and inefficiency, hindering the capture of rapidly changing physiological processes and biological phenomena. To address these problems, we propose an integrated intelligent sensor for biological fluorescence imaging and ultrafast recognition. This sensor integrates an imaging system based on a photodetector array and a recognition system based on neural networks on a single chip, featuring a highly compact structure, a continuously adjustable optical response, and reconfigurable electrical performance. The unified architecture of the imaging and recognition systems enables ultrafast recognition (19.63 μs) of tumor margins. Additionally, the special organic materials and bulk heterojunction structure endow the photodetector array with strong wavelength dependence, achieving high specific detectivity (3.06 × 10¹² Jones) in the narrowband near-infrared range commonly used in biomedical imaging (600–800 nm). After training, the sensor can accurately recognize biological fluorescence edges in real time, even under interference from other colored light noise. Benefiting from its rapidity and high accuracy, we demonstrated a simulated surgical experiment showcasing tumor edge fluorescence imaging, recognition, and cutting. This integrated approach holds the potential to establish a novel paradigm for designing and manufacturing intelligent medical sensors.

中文翻译：

---

用于集成生物荧光成像和识别的超快智能传感器


荧光成像和识别是靶向医学、病理外科和生物医学的核心技术。然而，当前的成像和识别系统是独立的，需要重复数据传输以进行成像和识别，这会导致延迟和效率低下，阻碍捕捉快速变化的生理过程和生物现象。为了解决这些问题，我们提出了一种用于生物荧光成像和超快速识别的集成智能传感器。该传感器在单个芯片上集成了基于光电探测器阵列的成像系统和基于神经网络的识别系统，具有高度紧凑的结构、连续可调的光学响应和可重新配置的电气性能。成像和识别系统的统一架构可实现肿瘤边缘的超快识别 （19.63 μs）。此外，特殊的有机材料和体异质结结构赋予光电探测器阵列强大的波长依赖性，在生物医学成像常用的窄带近红外范围 （3.06 ×^{10 12} Jones） （600-800 nm） 中实现了高特异性探测率 （600-800 nm）。经过训练，传感器可以实时准确识别生物荧光边缘，即使在其他有色光噪声的干扰下也是如此。得益于其快速性和高精度，我们演示了一项模拟手术实验，展示了肿瘤边缘荧光成像、识别和切割。这种集成方法有可能为设计和制造智能医疗传感器建立一种新的范式。