Machine Learning Aided Photonic Diagnostic System for Minimally Invasive Optically Guided Surgery in the Hepatoduodenal Area

Abstract

Abdominal cancer is a widely prevalent group of tumours with a high level of mortality if diagnosed at a late stage. Although the cancer death rates have in general declined over the past few decades, the mortality from tumours in the hepatoduodenal area has significantly increased in recent years. The broader use of minimal access surgery (MAS) for diagnostics and treatment can significantly improve the survival rate and quality of life of patients after surgery. This work aims to develop and characterise an appropriate technical implementation for tissue endogenous fluorescence (TEF) and assess the efficiency of machine learning methods for the real-time diagnosis of tumours in the hepatoduodenal area. In this paper, we present the results of the machine learning approach applied to the optically guided MAS. We have elaborated tissue fluorescence approach with a fibre-optic probe to record the TEF and blood perfusion parameters during MAS in patients with cancers in the hepatoduodenal area. The measurements from the laser Doppler flowmetry (LDF) channel were used as a sensor of the tissue vitality to reduce variability in TEF data. Also, we evaluated how the blood perfusion oscillations are changed in the tumour tissue. The evaluated amplitudes of the cardiac (0.6-1.6 Hz) and respiratory (0.2-0.6 Hz) oscillations was significantly higher in intact tissues (p < 0.001) compared to the cancerous ones, while the myogenic (0.2-0.06 Hz) oscillation did not demonstrate any statistically significant difference. Our results demonstrate that a fibre-optic TEF probe accompanied with ML algorithms such as k-Nearest Neighbours or AdaBoost is highly promising for the real-time in situ differentiation between cancerous and healthy tissues by detecting the information about the tissue type that is encoded in the fluorescence spectrum. Also, we show that the detection can be supplemented and enhanced by parallel collection and classification of blood perfusion oscillations.

Keywords: blood perfusion; endogenous fluorescence; laser Doppler flowmetry; liver cancer; machine learning; minimally invasive interventions.

Figure 1

The configuration and measuring channels of the used fibre-optic probe.

Figure 2

The components of measuring setup and probe placement in the region of interest.

Figure 3

Exemplary blood perfusion records (upper panel) from the intact bile duct wall and the tumour located in the bile duct and the corresponding integrated wavelet spectra (IWS; bottom panel).

Figure 4

5 arbitrary fluorescence spectra and their labels obtained at the excitation channel of 365 nm (left) and 450 nm (right). In these graphs, the label “N” denotes a normal tissue whereas “C” a cancerous tissue.

Figure 5

Example of percentage of explained variance at the excitation channel of 365 nm (left) and 450 nm (right) achieved by PCA.

Figure 6

Results on the statistical analysis (One-Way ANOVA with Tukey’s post hoc test, *** p < 0.001, n.s.: not significant, Mean ± SE) of the differences in the oscillations in normal intact and cancerous tissues in the hepatoduodenal area.