AI Detects Osteoporosis With High Accuracy By Analyzing Dental X-Rays

Scientists in Japan were able to identify osteoporosis with high accuracy using AI to analyze routine dental x-rays. The researchers at Kagawa Prefectural Central Hospital, Okayama University, and Matsumoto Dental University used deep learning to construct an osteoporosis classifier from dental x-rays. The study entitled Identification of osteoporosis using ensemble deep learning model with panoramic radiographs and clinical covariates was published in Scientific Reports on April 12, 2022.

General screening for osteoporosis using deep learning to analyze dental x-rays would be efficient and cost-effective and could help millions of people avoid fractures.

It's estimated that over 200 million people worldwide have osteoporosis. People with osteoporosis are at high risk for sudden bone fractures. About 50% of women and 30% of men will experience a fracture at some point in their lifetime. The number?of fractures?is expected to increase sharply due to worldwide population aging. Fracture rates vary between countries. For example, hip fracture rates are more common in Scandinavia and North America than in southern European, Asian and Latin American countries.

Fracture Risk Varies

Fracture risk for women worldwide. Source: International Osteoporosis Foundation

Fracture risk for men worldwide. Source: International Osteoporosis Foundation

Hip fracture rates after age adjustment. Image Source:?International Osteoporosis Foundation

Measuring Bone Density To Assess Risk

Dual-energy x-ray absorptiometry (DEXA) scans of the hip and spine are traditionally used to assess fracture risk and diagnose osteoporosis. Since bone density measurement using DEXA scans are relatively expensive, they are not suitable for general screening. Dental x-rays are ideal for general screening because they are low cost and can be performed by dentists during routine dental visits. If the scan detects loss of mineral density the dentist can refer the patient to an osteoporosis specialist.

Panoramic?x-ray is a type of x-ray that dentists use to capture the entire mouth including teeth, upper jaw, and lower jaw in one image. Dentists use this type of x-ray to prepare treatment for procedures such as dentures, braces, and extractions, and to identify?periodontal disease, impacted teeth, and wisdom teeth.

The researchers measured bone density on the mandible (the?lower jawbone). The mandible?is the largest and strongest bone in the human?facial skeleton. The researchers used deep learning to analyze the region of the lower jawbone marked in blue. The visualization of the feature area was different between ResNet and EfficientNet. This result is extremely interesting: ResNet focused on the cortical bone in the mandibular lower border but EfficientNet focused on the area above the cortical bone in addition to the cortical bone in the mandibular lower border.

The lower jaw looses mineral density as adult humans age. Image source: Wikipedia

Study Overview

1. In this study scientists developed an osteoporosis classifier based on an ensemble model.
2. The clinical covariates of patients were added to dental panoramic x-rays to statistically clarify the effect of classification accuracy on the addition of clinical covariates.
3. A dataset containing 778 images was collected from patients who underwent both skeletal-bone-mineral density measurement and dental panoramic x-rays at a single general hospital between 2014 and 2020.
4. In this study, the evaluation was performed using the standard convolutional neural network (CNN) models EfficientNet -b0, -b3, -b7 and ResNet-18, -50, -152.
5. ResNet is a CNN that won the classification task of the ILSVRC2015 Challenge. Generally, deepening the network layer improves the accuracy of image identification, but a network layer that is too deep reduces the accuracy.
6. EfficientNet is a CNN that was proposed as a state-of-the-art image classification method on ImageNet data. Although the number of parameters is smaller than that of the conventional CNN model, EfficientNet is a high-speed and relatively accurate CNN model.
7. Deep learning analysis was implemented using the PyTorch deep learning framework and Python.
8. An ensemble model was constructed with clinical covariates added to each CNN.
9. Since osteoporosis is associated with systemic patient factors the scientists hypothesized that the diagnostic accuracy using deep learning and X-ray images would be improved by constructing a CNN in which patient factors are added.

Image source: Identification of osteoporosis using ensemble deep learning model with panoramic radiographs and clinical covariates.

Findings and Conclusions

1. This study demonstrates that CNNs can diagnose osteoporosis from dental panoramic radiographs with high levels of accuracy.
2. The researchers showed that including patient variables involved in routine clinical settings improved the performance metrics of all predictions compared to using the image-only model.
3. The ensemble model was more effective for the CNN model with fewer parameters.
4. An ensemble model that included patient covariates demonstrated more accurate classification of osteoporosis.
5. The EfficientNet-B7 and ResNet-152 ensemble models showed the highest accuracy.
6. The visualization of the feature area was different between ResNet and EfficientNet. This result is extremely interesting - see visualization image above.
7. ResNet focused on the cortical bone in the mandibular lower border.
8. EfficientNet focused on the area above the cortical bone in addition to the cortical bone in the mandibular lower border.
9. This suggests that deep learning using X-ray images can be useful for diagnosing osteoporosis.
10. These results are expected to play an important role in the screening of osteoporosis in dental offices.

References

Identification of osteoporosis using ensemble deep learning model with panoramic radiographs and clinical covariates, Scientific Reports, April 12, 2022

Authors: Shintaro Sukegawa,?Ai Fujimura,?Akira Taguchi,?Norio Yamamoto,?Akira Kitamura,?Ryosuke Goto,?Keisuke Nakano,?Kiyofumi Takabatake,?Hotaka Kawai,?Hitoshi Nagatsuka, Yoshihiko Furuki?

Author Affiliations: Kagawa Prefectural Central Hospital (Kagawa, Japan), Okayama University (Okayama, Japan), Matsumoto Dental University, (Nagano, Japan).

Subscribe and Comment

I'm interested in your feedback - please leave your comments. To subscribe please click subscribe at the top of this article.

This article was written by?Margaretta Colangelo.?Margaretta is Co-founder of Jthereum. She serves on the advisory board of the AI Precision Health Institute at the University of Hawai?i?Cancer Center. She's based in San Francisco??@realmargaretta