Fast and Accurate Ophthalmic Medication Bottle Identification Using Deep Learning on a Smartphone Device.

PURPOSE: To assess the accuracy and efficacy of deep learning models, specifically convolutional neural networks (CNNs), to identify glaucoma medication bottles. DESIGN: Algorithm development for predicting ophthalmic medication bottles using a large mobile image-based dataset. PARTICIPANTS: A total of 3750 mobile images of 5 ophthalmic medication bottles were included: brimonidine tartrate, dorzolamide-timolol, latanoprost, prednisolone acetate, and moxifloxacin. METHODS: Seven CNN models were initially pretrained on a large-scale image database and subsequently retrained to classify 5 commonly prescribed topical ophthalmic medications using a training dataset of 2250 mobile-phone captured images. The retrained CNN models' accuracies were compared using k-fold cross-validation (k = 10). The top 2 performing CNN models were then embedded into separate iOS apps and evaluated using 1500 mobile images not included in the training dataset. MAIN OUTCOME MEASURES: Prediction accuracy, image processing time. RESULTS: Of the 7 CNN architectures, MobileNet v2 yielded the highest k-fold cross-validation accuracy of 0.974 (95% confidence interval [CI], 0.966-0.980) and the shortest average image processing time at 3.45 (95% CI, 3.13-3.77) sec/image. ResNet V2 had the second highest accuracy of 0.961 (95% CI, 0.952-0.969). When the 2 app-embedded CNNs were compared, in terms of accuracy, MobileNet V2, with an image prediction accuracy of 0.86 (95% CI, 0.84-0.88), was significantly greater than ResNet V2, 0.68 (95% CI, 0.66-0.71) (Table 1). Sensitivities and specificities varied between medications (Table 1). There was no significant difference in average imaging processing time, 0.32 (95% CI, 0.28-0.36) sec/image and 0.31 (95% CI, 0.29-0.33) sec/image for MobileNet V2 and ResNet V2, respectively. Information on beta-testing of the iOS app can be found here: https://lin.hs.uci.edu/research/. CONCLUSIONS: We have retrained MobileNet V2 to accurately identify ophthalmic medication bottles and demonstrated that this neural network can operate in a smartphone environment. This work serves as a proof-of-concept for the production of a CNN-based smartphone application to empower patients by decreasing risk for error.

The Relationship Between the Functional Gait Assessment and Quality-of-Life Data in Patients Undergoing Vestibular Schwannoma Resection.

OBJECTIVE: To examine the relationship between the Functional Gait Assessment (FGA) and quality of life (QOL) measurements relating to balance before and after vestibular schwannoma (VS) resection and to assess the role of preoperative FGA in predicting postoperative QOL. STUDY DESIGN: A prospective clinical study of adult patients undergoing VS resection between September 2018 and December 2019. FGA was administered 1 week before and after surgery. Dizziness Handicap Inventory (DHI) and Penn Acoustic Neuroma Quality of Life (PANQOL) were administered preoperatively and at 3 months postoperatively. SETTING: Single tertiary center. PATIENTS: Patients (age ≥ 18 years old) with VS undergoing microsurgical resection. Excluded were patient with previous surgery or radiation. INTERVENTION: VS resection. MAIN OUTCOMES AND MEASURES: Primary outcome: correlation between FGA and QOL surveys. Secondary outcome: correlation between preoperative measurements of balance and postoperative PANQOL. RESULTS: One hundred thirty-eight patients were analyzed (mean age: 48 years old, 65.9% female). The translabyrinthine approach was most commonly performed. Under multivariate analysis, preoperative FGA significantly correlated with preoperative PANQOL balance score (p < 0.0001), preoperative PANQOL total score (p = 0.0002), and preoperative DHI (p < 0.0001). However, postoperative FGA did not significantly correlate with postoperative PANQOL balance or total scores (p = 0.446 and p = 0.4, respectively), or postoperative DHI (p = 0.3). Univariate analysis demonstrated that preoperative DHI and preoperative FGA were predictive of changes in postoperative PANQOL balance and total scores. However under multivariate analysis, preoperative FGA did not predict changes in postoperative PANQOL balance or total score (p = 0.24; p = 0.28, respectively). Preoperative DHI remained predictive of changes in postoperative PANQOL balance (p = 0.03) score but not of postoperative PANQOL total score (p = 0.37). CONCLUSIONS: Although FGA and QOL data significantly correlated in the preoperative setting, our results did not suggest that preoperative FGA can be used to determine postoperative QOL. Additionally, the lack of correlation between FGA and QOL measurements in the acute postoperative setting suggests that further research is needed to determine contributors to postoperative QOL.