FP557

Axial Length Prediction by Artificial Intelligence Deploying Machine Learning Regression Algorithm

Aims: To find predictability equation and accuracy on training the model with keratometry values as input and A-scan derived AL 
Methods: Previous data of horizontal (Kh) and vertical curvatures (Kv) with axial length (AL) values collected from 108 eyes of 108 patients. Anaconda distribution for Python 3.7 and Jupyter Notebook downloaded and installed. After importing Pandas and numpy libraries, the data was pre-processed. The model is trained by LinearRegression function after importing train_test_split from sklearn portal. Seventy percent of dataset split for training and 30% for validation. Performance was validated by r2_score metrics extracted from the sklearn repository.
Results:The predicted model equation discovered was 0.39 x Kh + 0.10 x Kv + 0.94 and works for K values from 41.00 to 47.00 D and AL of 20 to 24 mm and the model achieved 0.97 (97%) accuracy.
Conclusion: This model helps in cross validating A-scan AL in doubtful cases and preventing post-op refractive surprise.