Deep Learning and LLM for Knee Osteoarthritis Grading on Low-Power Devices
A new research paper proposes an automated diagnostic approach for knee osteoarthritis (KOA) severity grading using a deep learning convolutional neural network (CNN) optimized for computationally limited systems. The model, based on ResNet-18, is trained on a publicly available database via transfer learning to classify knee images into five Kellgren-Lawrence (KL) grades. The approach integrates TensorFlow Lite for device-based inference, aiming to reduce subjectivity and inter-observer variability in conventional KOA diagnosis. The paper is published on arXiv with ID 2605.05731.
Key facts
- Knee osteoarthritis (KOA) is a musculoskeletal disorder causing chronic pain and reduced mobility.
- Conventional KOA diagnosis suffers from subjectivity and inter-observer variability.
- The proposed method uses a ResNet-18 CNN with transfer learning.
- The model classifies knee images into five Kellgren-Lawrence (KL) grades.
- TensorFlow Lite enables inference on computationally limited devices.
- The approach aims for precise and timely diagnosis.
- The model is trained on a publicly available database.
- The paper is available on arXiv with ID 2605.05731.
Entities
Institutions
- arXiv