Artificial Intelligence Based Autonomous Socket Proposal Program: Socket Design Experiences

Overview

The aim of this study is to develop an artificial intelligence-based autonomous socket recommendation program that will provide a more comfortable and easier test socket production with high time-cost efficiency and to share experiences about socket designs in these processes.

For the artificial intelligence-based software planned to be created, the stumps of all patients were scanned with the Artec Eva Lite brand 3D scanner. The scanned patterns were saved as point clouds. The socket parts of the prostheses used by the same patients were also scanned with the same scanner device and recorded. The point dataset consisting of stump-socket matches obtained from the patients was used for the software. In order to train the artificial intelligence model, a working environment has been created in which artificial intelligence libraries and tools can be used on the computer. For this purpose, first Anaconda data science platform was established. Thereupon, Python programming language and Tensorflow deep learning library were installed, other libraries required for the training of the artificial intelligence model were added, and the working environment was made ready. A deep learning algorithm was used in the artificial intelligence model developed for training the data. The purpose of using deep learning, which is one of the most up-to-date and popular artificial intelligence algorithms, is to achieve more accurate results by increasing the performance and accuracy rate. First, the dataset is 90% reserved for training and 10% for testing. Then, a deep learning model was created with the Sequantial() model selected from the Keras library. In the model, a total of 7 layers are used, the first of which is the input layer and the last is the output layer. While "relu" is used as the activation function for the input layer and intermediate layers, the "linear" function is used for the output layer. While creating the model, "Adam" was chosen as the optimizer. In the model trained with a total of 500 "repetitions", "batch size" is assigned as 5. The trained model was then tested with the test data and a success rate of 61% was achieved. Afterwards, the model and weights were recorded. After the model training was completed, a new Python program was developed. The previously developed models and weights were loaded while the program was running and were used to propose a socket for the new die data to be given. When the program is run, the stump name for which a socket is requested is asked. Thus, the program proposes a new socket after receiving the stubby data set from the user and testing it in the trained model. This 3D socket model is shown to the user via the Python Plotly Graphics Library.

Conditions

Interventions

Eligibility

Locations (1)

Outcomes