Diffuse reflectance and fluorescence spectroscopy for breast conserving surgery

Abstract

Purpose: The major challenge in breast conserving surgery is the high rates of re-excision due to positive resection margins. This study evaluates whether a combined diffuse reflectance spectroscopy (DRS) and laser induced intrinsic fluorescence spectroscopy (IFS) technique can differentiate breast tissue sample types, towards the development of an intraoperative margin assessment tool.

Methods: Breast tissue samples were collected from patients undergoing breast cancer surgery. A handheld DRS-IFS probe was used on the frozen thawed ex-vivo breast samples to acquire spectral data. Machine learning classifiers were used to determine sensitivity, specificity, overall diagnostic accuracy, and the area under the curve (AUC) against "gold-standard" histopathology ground truth.

Results: 181 breast tissue samples from 138 patients were interrogated using DRS-IFS. All patients were female, with median age (range) of 56.8 (20-94) years The total number of spectra acquired was 18,349. Following five-fold cross validation for normal versus cancer tissue, extreme gradient boost classifier achieved a sensitivity of 84% (SD ± 13), specificity of 61% (SD ± 16), overall diagnostic accuracy of 75% (SD ± 3), and AUC of 84%.

Conclusion: The results suggests that DRS-IFS can distinguish normal breast tissue from breast cancer with high diagnostic accuracy. For DRS-IFS to be translated into the operating theatre to aid a surgeon's real-time visualisation for oncologic margin control assessment of intraoperative, the in vivo diagnostic accuracy needs to be determined.

Keywords: Breast cancer, margins; Diffuse reflectance spectroscopy; Intraoperative margin assessment, spectroscopy; Intrinsic fluorescence.

Fig. 1

DRS-IFS experimental setup: a Schematic illustration depicting the components of the custom-built experimental system. b Diagram of configuration of the probe components and distances shown in mm. c Sample holder and fibre-optic probe with tissue sample on top and fibre probe in contact with bottom surface of the sample. The spatial resolution of the system is 3–5 mm. The DRS detection volume varies from approximately 0.5–1 mm³ at a 0.5 mm fibre separation to several mm³ at 2.8 mm fibre separation, and interrogation depth varies from 0.1 mm up to 3.5 mm, although this will be affected by the specific optical properties of each breast tissue sample. For IFS, the detection volume is in order of 1 mm², the interrogation depth is less than 0.5 mm. d Photograph of the DRS-IFS multifibre probe distal tip, inset – photograph of the fibre configuration

Fig. 2

Illustration of the four different acquisition modes of the DRS-IFS system representing DRS and IFS data collection. DRS a) and b) sampling volumes are approximately represented by the coloured areas, and were estimated via a cloud-based Monte Carlo simulation tool [21] for human fatty breast tissue with absorption coefficient 0.041 cm⁻¹ and reduced scattering coefficient 8.5 cm⁻¹ for DRS at 786 nm[22]. a demonstrates simultaneous acquisition of DRS spectra at distances 0.5 and 2.8 mm, b demonstrates simultaneous acquisition of DRS spectra at distances 0.8 mm and 1.6 mm, c and d demonstrate acquisition of IFS spectra at 375 nm and 405 nm excitation. The tissue depth is depicted by “Z” on the left-hand side and represented in millimetres

Fig. 3

a Macroscopic photograph of breast tissue sample interrogated with the DRS-IFS system. b Corresponding histopathology slide image following review by an experienced consultant pathologist identified invasive lobular cancer. c Macroscopic photograph of a normal breast tissue sample with healthy appearance. d Corresponding histopathology slide identifying healthy breast tissue

Fig. 4

Summary of the main steps involved in preprocessing and classification of the acquired spectra. The classification was repeated for different combinations of the acquired spectra

Fig. 5

Graphs depicting mean spectra and standard deviations (represented by vertical bars) from each individual channel after preprocessing of data from all tissue samples regardless of quality: fluorescence excitation at (a) 375 nm, (b) 405 nm (c–f) spectral features from DRS at (c) 0.5 mm, (d) 0.8 mm, (e) 1.6 mm, (f) 2.8 mm. DRS diffuse reflectance spectroscopy, IFS intrinsic fluorescence spectroscopy

Fig. 6

Confusion matrix comparing healthy tissue samples versus all quality one (good) tumour samples (IDC & ILC & DCIS) for the channel combination of DRS 0.8 mm, DRS 2.8 mm, and IFS 405 nm, once the XGB classifier is applied

Fig. 7

Confusion matrix comparing healthy tissue samples versus all tumour samples (IDC & ILC & DCIS) irrespective of quality score for the channel combination of DRS 0.8 mm, DRS 2.8 mm, and IFS 405 nm and Fixed Windows Features