Millward Steven W, Wei Peng, Piwnica-Worms David, Gammon Seth T
Department of Cancer Systems Imaging, The University of Texas MD Anderson Cancer Center, Houston, TX 77030, USA.
Department of Biostatistics, The University of Texas MD Anderson Cancer Center, Houston, TX 77030, USA.
Cancers (Basel). 2025 Apr 22;17(9):1387. doi: 10.3390/cancers17091387.
Over the past 30 years, academic and industrial research investigators have developed molecular reporters to visualize cell death in complex biological systems. In parallel, clinical researchers, chemists, biochemists, and molecular biologists have endeavored to translate these molecular tools into clinical imaging agents. Despite these efforts, there are no clinically approved imaging methodologies with which to image cell death consistently and quantitatively. One reason may reside in the intrinsic mismatch between the sampling frequency of translational molecular imaging and the biochemical kinetics that define cell death. Beyond cell death imaging, many active research programs are now attempting to create translational diagnostic pharmaceuticals to image immunological, fibrotic, amyloidotic, and metabolic pathways. Each of these pathways is defined by a unique set of biochemical rate constants, some of which are associated with key predictive pathways. Exhaustively sampling all permutations of pathways and kinetic constants would seem to be an intractable strategy for target identification and validation. Sampling theory, if applied to these pathways, could accelerate the translation of high-impact diagnostics through prioritization of pathways for either AI enhanced diagnostic imaging or AI-enhanced wearable devices. In this perspective, we identify the Nyquist sampling rate as a key criterion for evaluating the optimal application for novel diagnostics. Sampling theory states that to fully characterize a band-limited, stationary, temporal data set, the signal must be sampled at more than twice the rate of the fastest frequency in the signal or, for diagnostics, the discriminatory signal. Through the study of the medical imaging process chain, Nyquist sampling rates of 0.25 day and, more likely, slower than 0.02 day were determined to provide high quality information. By prioritizing low-frequency predictive processes, or "state changes,", imaging researchers may improve the "hit rate" of research programs by appropriately matching the rate of change in diagnostic and predictive information with the limiting sampling rate of medical imaging. Critically, however, high-frequency diagnostic information (and therefore high-frequency biological processes) need not be ignored; these processes are simply better interrogated through continuous monitoring, e.g., by wearable devices combined with machine learning or artificial intelligence.
在过去30年里，学术和产业研究人员开发了分子报告基因，以在复杂生物系统中可视化细胞死亡。与此同时，临床研究人员、化学家、生物化学家和分子生物学家一直致力于将这些分子工具转化为临床成像剂。尽管做出了这些努力，但目前尚无临床批准的成像方法能够持续且定量地对细胞死亡进行成像。一个原因可能在于转化分子成像的采样频率与定义细胞死亡的生化动力学之间存在内在不匹配。除了细胞死亡成像，许多活跃的研究项目现在正试图创建转化诊断药物，以对免疫、纤维化、淀粉样变性和代谢途径进行成像。这些途径中的每一个都由一组独特的生化速率常数定义，其中一些与关键预测途径相关。详尽地对途径和动力学常数的所有排列进行采样，对于靶点识别和验证而言似乎是一种难以处理的策略。如果将采样理论应用于这些途径，通过对人工智能增强诊断成像或人工智能增强可穿戴设备的途径进行优先级排序，可以加速高影响力诊断方法的转化。从这个角度来看，我们将奈奎斯特采样率确定为评估新型诊断方法最佳应用的关键标准。采样理论指出，要全面表征一个带限、平稳的时间数据集，信号的采样速率必须高于信号中最快频率的两倍，或者对于诊断而言，高于鉴别信号的两倍。通过对医学成像过程链的研究，确定0.25天以及更可能低于0.02天的奈奎斯特采样率能够提供高质量信息。通过对低频预测过程（即“状态变化”）进行优先级排序，成像研究人员可以通过使诊断和预测信息的变化速率与医学成像的极限采样率适当匹配，来提高研究项目的“命中率”。然而，至关重要的是，高频诊断信息（以及因此的高频生物过程）不必被忽视；这些过程通过连续监测，例如通过与机器学习或人工智能相结合的可穿戴设备，能得到更好的研究。
