Diagnosis of Lumbar Spondylolisthesis Using Optimized Pretrained CNN Models.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Spondylolisthesis, the slippage of one vertebra over another, is a chronic condition necessitating early detection to avoid surgery.
• Accurate and timely diagnosis of spondylolisthesis is crucial for effective patient management and treatment planning.

Purpose of the Study:

• To develop and optimize a deep learning model for the automated detection of spondylolisthesis in X-ray radiographs.
• To enhance the efficiency and accessibility of spondylolisthesis diagnosis using optimized AI models suitable for mobile devices.

Main Methods:

• Utilized a dataset of 299 X-ray radiographs (156 with spondylolisthesis, 143 normal).
• Employed image augmentation to expand the dataset and VGG16 and InceptionV3 models for image classification.
• Applied TFLite model optimization techniques to reduce model size and improve performance.

Main Results:

• The VGG16 model achieved 98% accuracy, surpassing InceptionV3's 96% accuracy before optimization.
• Optimized (quantized) VGG16 and InceptionV3 models reached 100% and 96% accuracy, respectively.
• The optimized models demonstrated superior performance compared to existing methods and were reduced in size for potential use on mobile devices.

Conclusions:

• Optimized deep learning models, particularly VGG16, show significant promise for accurate and efficient spondylolisthesis detection from X-rays.
• The developed models generalize well across different datasets, indicating a robust approach for widespread clinical application.
• Model optimization using TFLite facilitates deployment on resource-constrained devices, potentially improving diagnostic accessibility.