RESUMEN
Objective:To explore whether deep learning could apply to recognize the recurrent laryngeal nerve (RLN) in videos of endoscopic thyroidectomy (ETE) via breast approach.Methods:Videos of ETE via breast approach in Peking Union Medical College Hospital from Feb. 2020 to Aug. 2021 were collected. Videos containing RLN were selected, and the outline of RLN was marked by two thyroid surgeons. Then data were divided into a training set and a test set in a ratio of 5:1 and classified into the high and low difficulty group according to a senior thyroid surgeon’s opinion. Those pictures were input to D-LinkNet model. Precision, sensitivity and mean dice index was calculated.Results:A total of 46 videos including 153, 520 frames of pictures were included in this study. 131,039 frames of 39 videos were in the training set and 22,481 frames of 7 videos were in the test set. When the intersection over union threshold was 0.1, the sensitivity and precision was 92.9%/72.8% and 47.6%/54.9% in high and low recognition group, respectively. When the intersection over union threshold was 0.5, the sensitivity and precision turned to 85.8%/67.2% and 37.6%/43.5% in high and low difficulty group, respectively. Mean Dice index was 0.781 and 0.663 in high and low difficulty group, respectively.Conclusions:RLN recognition based on deep learning is feasible and has potential application value in ETE, which may help surgeons reduce the risk of accidental injury of RLN and improve the safety of thyroidectomy.
RESUMEN
Objective:To explore whether deep learning could apply to recognize the recurrent laryngeal nerve in the video of unilateral axillary approach endoscopic thyroidectomy.Methods:Videos of endoscopic thyroidectomy via unilateral axillary approach in Peking Union Medical College Hospital from Jul. 1st, 2020 to May. 1st, 2021 were collected. Videos containing the recurrent laryngeal nerve were selected, and the outline of recurrent laryngeal nerve were marked by two senior thyroid surgeons and staffs. Data were divided into training set and test set in a ratio of 5:1, and classified into high, medium and low recognition group according to difficulty of recognizing the outline of the nerve. The neuron network was based on PSPNet combined with Resnet50. All data were analyzed by R (ver. 4.0.2) .Results:A total of 38 videos including 35,501 frames of pictures were included in this study. 29, 704 frames of 32 videos were in our training set and 5797 frames of 6 videos were in the test set. When the intersection over union (IOU) threshold is 0.1, the sensitivity and precision is 100.0%/92.1%, 95.8%/80.2% and 81.0%/80.6% in high, medium and low recognition group respectively. When the IOU threshold is 0.5, the sensitivity and precision is 92.6%/85.3%, 71.7%/60.5% and 38.1%/37.9% in high, medium and low recognition group respectively, indicating that neuron network could located the outline of recurrent laryngeal nerve in high and medium recognition group. False negatives were often due to small targets and unclear boundaries.Conclusion:Recurrent laryngeal nerve recognition based on deep learning is feasible and has potential application value in endoscopic thyroidectomy, which may help surgeons reduce the risk of accidental injury of recurrent laryngeal nerve and improve the safety of thyroidectomy.