Intraoperative Ex Vivo Pituitary Adenoma Subtype Classification Using Noncontact Laser Fluorescence Spectroscopy

Background: Near real-time intraoperative identification of tumor tissue is needed, as contemporary methods such as intraoperative MRI are expensive, time intensive, and only available at the most advanced care facilities. The TumorID is a low cost, portable, non-contact laser induced fluorescence spectroscopy device designed by our research group for intraoperative use. Importantly, the TumorID uses autofluorescence, meaning no additional fluorophores are given to the patient thereby avoiding the associated risks and timeline constraints. By using an excitation wavelength of 405 nm the TumorID targets differences in NADH and FAD noted by the Warburg effect. Providing the surgical team with increased knowledge about tumor location could potentially increase resection, decrease surgical time, and therefore increase patient outcomes. This study shows the TumorID's applicability in identifying differences using a logistic regression model in recently resected Pituitary Adenoma with an endonasal approach.

Methods: After the tissue was resected and the appropriate samples were sent to pathology, the remaining tissue was placed under the TumorID ([Fig. 1]) directly off the sterile field for spectral collection before going to pathology. As many points as possible spaced 1 mm away from each other, determined by the laser spot size, were collected with a laser power of 180 mW and an integration time of 0.5 s. This laser power does not alter the tissue. Tissue samples were collected from non-secreting pituitary adenoma (n = 5) and prolactin secreting pituitary adenoma (n = 4) cases. The data were normalized with respect to the largest intensity value greater than 425 nm, the dichroic filter's cutoff wavelength. The data were then smoothed using a moving average to decrease the signal to noise ratio. At this point, the data were split into training (70%) and testing (30%) classes to fit a standard logistic regression model.

Results: The model had a total accuracy of 93% when tested on 14 scans. Since it is a multiclass classification model, the precision, recall, and F1 scores for each class are provided in [Table 1].

Table 1 Model performance statistics

	Non secreting	Prolactin secreting
Precision	80%	100%
Recall	100%	90%
F1 score	89%	95%

The data can be seen with confidence intervals calculated at each of the 3,648 wavelengths and a visual difference between the two tissue types can be seen ([Fig. 2]).

Conclusion: The results of this study support the hypothesis that the TumorID can provide quick, accurate classification of pituitary adenoma tissue in the operating room. Further studies with healthy tissue are needed to determine if the TumorID can classify tumor tissue from healthy tissue; however, this study shows the effective application of non-contact fluorescence spectroscopy.

Fig. 1 TumorID device in the operating room.

Fig. 2 Mean spectral data with 95% confidence interval.

Fig. 3 Logistic regression classification matrix (percentage).

No conflict of interest has been declared by the author(s).

Publication History

Article published online:
01 February 2023