Near-infrared indocyanine green (ICG) angiography allows experienced surgeons reliably discern parathyroid gland (PG) vitality during thyroid and parathyroid operations and so predict post-operative function. To facilitate all surgeons perform similarly, we developed an automatic computational quantification method using computer vision that accurately portrays expert interpretation of visualised PG ICG angiographic fluorescence signals.

ICG-PG angiography video recordings (Fluobeam® LX, Fluoptics, Grenoble - part of Getinge - Göteborg) from patients undergoing endocrine cervical surgery in a high-volume unit were used developmentally. Computation (Matlab, Mathworks, Ireland) included segmentation-identification of PG (by autofluorescence), image stabilisation (by linear translation) and adjusted time-fluorescence intensity profile (TFIP) generation. Relative upslope and maximum intensity ratios then trained a simple logistic regression model grounded in expert interpretation and outcome (including hypoparathyroidism), with unseen testing subsequently.

The model was trained on 37 patient videos (45 glands, 29 judged well perfused by PG angiography experts), achieving feature data separation with 100% accuracy, and tested on 22 unseen videos (27 glands, 15 judged well perfused), including four live in-theatre. Segmentation-guided PG detection correctly identified all PGs during unseen testing along with three additional non-PG regions (90% positive predictive value). Subsequent TFIP extraction with vitality prediction proved feasible in all cases within five minutes, with 96.3% model prediction accuracy (sensitivity/specificity = 93.3/100% respectively) versus expert judgement.

Automatic PG perfusion quantification with simple machine learning computational methods discriminates PG perfusion concordant with expert surgeon interpretation, providing too a means for ICG-PG signal documentation.