Volume 52, Number 7, 1998 APPLIED SPECTROSCOPY 943 0003-7028 / 98 / 5207-0943$2.00 / 0 q 1998 Society for Applied Spectroscopy Method to Determine Tissue Fluorescence Ef®ciency in Vivo and Predict Signal-to-Noise Ratio for Spectrometers E. V. TRUJILLO, D. R. SANDISON, U. UTZINGER, N. RAMANUJAM, M. FOLLEN MITCHELL, and R. RICHARDS-KORTUM * Department of Electrical and Computer Engineering, The University of Texas, Austin, Texas, 78712. (E.V.T., U.U., N.R. and R.R-K.); Department 2665, Sandia National Laboratories, Albuquerque, New Mexico, 87185 (D.R.S.); and Department of Gynecologic Oncology, UT MD Anderson Cancer Center, Houston, Texas, 77035 (M.F.M.) Recent clinical trials have demonstrated the potential of ¯uores- cence spectroscopy for in vivo diagnosis of pathology. There is sig- ni®cant potential to reduce the cost and complexity of instrumen- tation to measure tissue spectra; however, careful analysis is re- quired to maximize performance and minimize cost. One measure of performance is the signal-to-noise ratio (SNR) of the resulting data. This paper describes a method to predict the SNR of a given optical design for a particular tissue application. In order to cal- culate the expected SNR, two pieces of information are required: (1) the throughput and inherent noise of the system and (2) a quan- titative relationship between the illumination energy and the re- sulting tissue ¯uorescence available for collection, which we de®ne as the tissue ¯uorescence ef®ciency (FE). We present a method to calculate the ¯uorescence ef®ciency of tissue from in vivo measure- ments of tissue ¯uorescence. We report FE measurements of the normal and precancerous human cervix in vivo at 337, 380, and 460 nm excitation. We also present and evaluate a method to estimate the throughput and noise of various spectrometers and predict the expected SNR for tissue spectra by using the measured tissue FE. For squamous cervical tissue, as the degree of the disease increases, FE decreases, and as the excitation wavelength increases, FE de- creases. Cervical tissue FE varies more than two orders of magni- tude, depending on the tissue type and on the excitation wavelength used. Our SNR calculations, based on measured values of tissue FE, demonstrate agreement within a factor of 1.3 of the measured SNR on average. This method can be used to estimate the performance of different spectrometer designs for clinical use. Index Headings: Cost-effectiveness analysis; Cervix; Precancer; Fluorescence spectroscopy; Signal-to-noise ratio. INTRODUCTION Recent clinical trials have demonstrated the potential of ¯uorescence spectroscopy for in vivo diagnosis of pa- thology. 1,2 Tissue ¯uorescence spectroscopy can be used to image large areas of tissue to identify areas suspicious for disease and extract diagnostically relevant structural and histochemical information. 3 This is an important ad- vantage with respect to many traditional methods that require clinical expertise to identify areas to biopsy for subsequent histologic analysis. In particular, ¯uorescence has shown promise for the diagnosis of precancerous and cancerous lesions of the breast, 4 lung, 5 bronchus, 6 oral cavity, 7 cervix, 8 gastrointestinal tract, 1 ,9 and brain. 10 Although spectroscopic measurements offer several advantages, instrumentation described in the literature is expensive relative to current diagnostic tools, and, in our experience, dif®cult for clinical staff to use without tech- nical assistance. There is signi®cant potential to reduce Received 3 November 1997; accepted 3 March 1998. * Author to whom correspondence should be sent. the cost and complexity of this instrumentation; however, careful analysis is required to maximize performance and minimize cost of instrumentation. One measure of the performance of an optical system is the signal-to-noise ratio (SNR) of the resulting data. In general, as the SNR is reduced, the classi®cation accu- racy dropsÐas do the system costs. 11±13 This paper de- scribes a method to predict the SNR of a given optical design for a particular tissue application and should serve as a guide in the design and evaluation of more cost- effective ¯uorescence diagnostic systems. We illustrate and validate this method with the particular example of detection of cervical precancer. METHODS In order to calculate the expected SNR for a given system design, two pieces of information are required: (1) the throughput and inherent noise of the system and (2) a quantitative relationship between the illumination energy and the resulting tissue ¯uorescence available for collection. In optically dilute samples, the quantum ef®- ciency (QE) provides this relationship, and methods to measure QE are well known. 3 However, in turbid, mul- ticomponent samples (such as tissue), these methods are not appropriate. To remove these limitations, we have developed a method to calculate the ¯uorescence ef®- ciency (FE) of tissue from in vivo measurements of tissue ¯uorescence. We report ¯uorescence ef®ciency measure- ments of the normal and precancerous human cervix in vivo at three excitation wavelengths. We also present a method to estimate the throughput and noise of various spectrometers and predict the expected SNR for tissue spectra by using the measured ¯uorescence ef®ciency of tissue. We validate this method for cervical tissue ¯uo- rescence spectra. FE Theory. In a dilute, homogeneous, nonscattering solution of pathlength L, Eq. 1 describes the relationship between the excitation energy ( P 0 ( l x )) and the ¯uores- cence energy arriving at the detector ( P ( l x , l m )), as a function of excitation and emission wavelength. L P ( l , l ) 5 dz P ( l ) m ( l ) f( l ) b(z). (1) x m E 0 x a x m 0 In this formula m a ( l x ) represents the absorption coef- ®cient of the ¯uorophore at the excitation wavelength, f( l m ) is the fraction of absorbed energy converted to ¯uorescence at the emission wavelength, and b(z) is the collection ef®ciency of the system, which depends on the