Using 2-dimensional hand photographs to predict postoperative biochemical remission in acromegaly patients: a transfer learning approach

Background: The primary treatment goals in acromegaly patients are complete surgical removal of underlying pituitary tumors and biochemical remission. One of the challenges in developing countries is the difficulty in monitoring postoperative biochemical levels in acromegaly patients, particularly those who live in remote areas or regions with limited medical resources. Methods: In an attempt to overcome the abovementioned challenges, we conducted a retrospective study and established a mobile and low-cost method to predict biochemical remission in acromegaly patients after surgery, the efficacy of which was assessed retrospectively using the China Acromegaly Patient Association (CAPA) database. A total of 368 surgical patients from the CAPA database were successfully followed up to obtain their hand photographs. Demographics, baseline clinical characteristics, pituitary tumor features, and treatment details were collated. Postoperative outcome, defined as biochemical remission at the last follow-up timepoint, was assessed. Transfer learning with a new mobile tailored neurocomputing architecture, MobileNetv2, was used to explore the identical features that could be used as predictors of long-term biochemical remission after surgery. Results: As expected, the MobileNetv2-based transfer learning algorithm was shown to predict biochemical remission with statistical accuracies of 0.96 and 0.76 in the training cohort (n=803) and validation cohort (n=200), respectively, and the loss function value was 0.82. Conclusions: Our findings demonstrate the potential of the MobileNetv2-based transfer learning algorithm in predicting biochemical remission for postoperative patients who are at home or live far away from a pituitary or neuroendocrinological treatment center.

Automated summary

This study presents a mobile and low-cost method to predict biochemical remission in acromegaly patients after surgery. The results show that the MobileNetv2-based transfer learning algorithm can accurately predict the likelihood of biochemical remission in acromegaly patients, providing a potential solution for those who live in remote areas or regions with limited medical resources.