摘要： Broadband mid-infrared spectroscopy of biofluids carries great potential for biological and biomedical applications, as it provides fast, reliable and label-free access to the molecular composition of the sample. When applied to human blood serum, a range of molecular contents can be quantified and specific changes in the absorption spectra, driven by diseases (e.g. cancer) can be identified and used for diagnostic purposes. One remaining challenge is the complexity of human serum: physiological phenotypes are driven by minor changes in concentration of thousands of different molecules. At the same time, although many low-abundance molecules are very informative for disease detection, these are often not detectable with conventional Fourier-Transform Infrared (FTIR) spectroscopy and quantum-cascade laser (QCL) based approaches due to a lack of sensitivity and specificity. Here we show how field-resolved spectroscopy (FRS) of few-cycle-excited molecular vibrations can be utilised to address these shortcomings and demonstrate its applicability for the measurement of human blood serum in a clinical setting. In a preparatory experiment, we quantitatively investigate the ability of FRS to detect small changes in the sample response by spiking blood serum with a defined concentration of dimethyl sulfone (DMSO2). We demonstrate that FRS is able to detect changes in molecular concentration down to the sub-μg/ml level in human blood serum, outperforming the sensitivity of FTIR- and QCL-based approaches. Hence, the smallest changes currently detectable by FRS are five orders of magnitude below the concentration of the most highly-abundant molecules in blood, implying a detectable concentration dynamic range of 105. Based on these results, we apply FRS for the first time in a real-world setting, a clinical study with a well-matched cohort of 195 control individuals and 58 lung, 41 prostate and 42 breast cancer patients and compare the obtained results to state-of-the-art FTIR measurements of the same samples. Corresponding infrared molecular fingerprints were recorded as time-domain sampled field oscillations emerging from the excited molecules in the serum samples (Fig. 1a). We find that investigation at different time windows (Fig. 1c-d) reveals a distinct temporal response of the samples originating from controls as compared to lung cancer patients. Preliminary analysis of the time-domain data with a random forest classifier reveals a high detection accuracy for cancer detection. Even though FRS, in this early stage of development, has considerably smaller spectral coverage as compared to FTIR, it reaches a similar rate of cancer detection efficiency. This suggests that the next generation FRS, with several octave spectral coverage, holds promise for outperforming FTIR fingerprinting substantially.