F.A. Koleda1, M.N. Belsheva2, A.A. Baimuratov3, L.P. Safonova4

1–4 Bauman Moscow State Technical University (Moscow, Russia)
1 koledafa@bmstu.ru, 2 belsheva@bmstu.ru, 3 baimuratov@bmstu.ru, 4 lpsafonova@bmstu.ru

Early detection of blood supply issues during tissue flap transplantation is crucial for determining the survival rate. Frequency domain near infrared spectroscopy (FD NIRS) enables the quantitative assessment of oxyhemoglobin, deoxyhemoglobin, and water concentrations. By recording the amplitude and phase of modulated light at frequencies between 50 and 1000 MHz, this method can also evaluate blood flow and oxygen consumption. These capabilities make FD NIRS a promising approach for intraoperative and postoperative monitoring of tissue flaps. However, the reproducibility of measurements depends on precise positional and force control. Multimodal acquisition of force and optical signals at multiple wavelengths and source-detector distances produces large data volumes, necessitating the use of field-programmable gate array (FPGA) for parallel processing. This study aims to develop a hardware-software complex (HSC) that accelerates the processing of narrowband FD NIRS data for tissue flap monitoring.

The developed system, based on FPGA, performs parallel processing of photodetector and reference (which does not pass through biotissue) signals. By calculating signal amplitude and phase using the discrete Fourier transform (DFT), it reduces both the amount of transmitted data and the computational load through preliminary data compression. The HSC operates within a developed heterodyne spectrometer featuring source modulation and cross-correlation frequencies of 110 MHz and 25 kHz, respectively. A 12-bit analog-to-digital converter with a 2.25 MHz sampling rate records the signals, providing 90 samples per 25 kHz signal period. To improve spectral analysis accuracy, four periods of the signal (360 samples points) are recorded for each laser diode. The HSC includes modules for DFT, amplitude and phase extraction, phase difference computation, averaging, and UART-based data exchange. Two parallel DFT computation pipelines process the photodetector and reference signal data. A single harmonic component is calculated using a systolic array combined with the CORDIC algorithm. Since further computation of optical and physiological parameters requires the photodetector signal amplitude and the phase difference between the photodetector and reference signals, the HSC incorporates a dedicated phase-difference module. To suppress noise, amplitude and phase-difference values are averaged for each channel corresponding to a single active laser diode. An external address counter controls the sequential output of averaged data, providing architectural flexibility and enabling channel reconfiguration without hardware redesign. An address counter located outside the module controls the sequential output of the averaged data, ensuring flexibility of the architecture and the ability to change the number of channels without reworking the system. The architecture supports both sequential and arbitrary diode multiplexing, allowing flexible selection of source-detector distances and adjustment of probing depth. Data exchange with external devices is performed via a UART interface that includes a phase-shifted reception module, improving tolerance to transmission rate deviations, and a buffered transmission module capable of sending 32-bit amplitude and phase values.

The developed HSC was tested using a heterodyne FD spectrometer and Saylinx development board based on Altera Cyclone IV FPGA. Signals were recorded from silicone tissue-mimicking phantoms with known optical properties, manufactured by ISS (USA), and from the hands of three volunteers during an occlusion test. Amplitude and phase errors were determined by processing recorded data on a computer and comparing the results with those obtained from the HSC. Relative amplitude errors and absolute phase errors were evaluated across 1,598 test samples. The HSC demonstrated amplitude detection with a deviation of less than 0.1% and phase detection with a deviation of 0.1834°. The CORDIC algorithm used 10 iterations, and the averaging count was set to 1. In future, averaging will be performed over eight samples ensuring a sampling frequency of 130 Hz. The corresponding amplitude deviation yields accuracy sufficient to determine optical parameters within 3%, adequate for real-time differential diagnosis of pathological conditions. Phase accuracy can be further improved by implementing the averaging module. The current tests were performed without averaging, as the FPGA has not yet been integrated into the spectrophotometer model with the sampling frequency 10 Hz. In future implementations, phase accuracy may be enhanced by adjusting the CORDIC algorithm parameters, increasing numerical precision, and using an FPGA with a larger logic cell capacity.

RAM_w_im:ram_im – RAM memory block for storing Fourier coefficients for calculating the imaginary part of the transform result; RAM_w_re:ram_re – RAM memory block for storing Fourier coefficients for calculating the real part of the transform result; serial_fft_coral:fft_core – block for calculating one frequency component of the discrete Fourier transform of the signal from the photodetector and the reference signal; w_im – weight input for calculating the imaginary part; w_re – weight input for calculating the real part; clk – clock input; x[0] – photodetector signal samples input; x[1] – reference signal samples input; valid_i – data validity input from the photodetector and reference signal; rstn – inverted reset input; im[0] – output of the imaginary part of the Fourier transform result of the photodetector signal; im[1] – output of the imaginary part of the Fourier transform result of the reference signal; re[0] – output of the real part of the Fourier transform result of the photodetector signal; re[1] – output of the real part of the Fourier transform result of the reference signal; counter – counter output; finish – data reception completion output; valid_o – data validity signal output

Koleda F.A., Belsheva M.N., Baimuratov A.A., Safonova L.P. FPGA-based frequency domain spectroscopy data processing for tissue flap monitoring. Biomedicine Radioengineering. 2025. V. 28. № 7. P. 76–87. DOI: https:// doi.org/10.18127/ j15604136-202507-08 (In Russian)

1. Rogoń I. et.al. Flap Monitoring Techniques: A Review. Journal of Clinical Medicine. 2024. V. 13. № 18.
2. Repež A., Oroszy D., Arnež Z.M. Continuous postoperative monitoring of cutaneous free flaps using near infrared spectroscopy. Journal of Plastic, Reconstructive & Aesthetic Surgery. 2008. V. 61. № 1. P. 71–77.
3. Yang J.C.-S. et al. The Use of Radial Vessel Stump in Free Radial Forearm Flap as Flap Monitor in Head and Neck Reconstruction. Annals of Plastic Surgery. 2007. V. 59. № 4.
4. Pohlenz P. et al. Microvascular free flaps in head and neck surgery: complications and outcome of 1000 flaps. International Journal of Oral and Maxillofacial Surgery. 2012. V. 41. № 6. P. 739–743.
5. Lohman R.F. et al. An Analysis of Current Techniques Used for Intraoperative Flap Evaluation. Annals of Plastic Surgery. 2015. V. 75. № 6. P. 679–685.
6. Fantini S., Sassaroli A. Frequency-Domain Techniques for Cerebral and Functional Near-Infrared Spectroscopy. Frontiers in neuroscience. 2020. V. 14. P. 300.
7. US5497769A. Photosensor with multiple light sources. Enrico Gratton, Sergio Fantini, Maria A. Franceschini, William Mantulin, Beniamino Barbieri. 1996.
8. Yamada Y., Suzuki H., Yamashita Y. Time-Domain Near-Infrared Spectroscopy and Imaging: A Review. Applied Sciences. 2019. V. 9. № 6. P. 1127.
9. Tuchin V.V. Opticheskaya medicinskaya diagnostika. T.1. (In Russian)
10. Zhang X.U. et al. Effect of probe pressure on skin tissue optical properties measurement using multi-diameter single fiber reflectance spectroscopy. Journal of Physics: Photonics. IOP Publishing. 2020. V. 2. № 3. P. 034008.
11. Giaconia G.C. et al. Exploring FPGA‐Based Lock‐In Techniques for Brain Monitoring Applications. Electronics. 2017. V. 6. № 1.
12. Kitsmiller V.J. Next Generation Frequency Domain Near Infrared Spectroscopy. University of Notre Dame. 2020.
13. Kita D.M. et al. High-performance and scalable on-chip digital Fourier transform spectroscopy. Nature Communications. 2018. V. 9. № 1. P. 4405.
14. Johnson K.T., Hurson A.R., Shirazi B. General-purpose systolic arrays. Computer. 1993. V. 26. № 11. P. 20–31.
15. Volder J.E. The CORDIC Trigonometric Computing Technique. IRE Transactions on Electronic Computers. 1959. V. EC-8. № 3. P. 330–334.
16. Nanda U., Pattnaik S.K. Universal Asynchronous Receiver and Transmitter (UART) // 2016 3rd International Conference on Advanced Computing and Communication Systems (ICACCS). 2016. Т. 01. С. 1–5.
17. Koch S., Sanders P., Williams M. BlockFIFO & MultiFIFO: Scalable Relaxed Queues: arXiv:2507.22764. arXiv. 2025.
18. Chance B. et al. Phase measurement of light absorption and scatter in human tissue. Review of Scientific Instruments. 1998 V. 69. № 10. P. 3457–3481.
19. Sthalekar C.C., Koomson V.J. A CMOS Sensor for measurement of cerebral optical coefficients using non-invasive frequency domain near infrared spectroscopy. In: IEEE Sensors Journal. V. 13. № 9. P. 3166–3174.