- Tytuł:
- A supervised approach to musculoskeletal imaging fracture detection and classification using deep learning algorithms
- Autorzy:
-
Karanam, Santoshachandra Rao
Srinivas, Y.
Chakravarty, S. - Powiązania:
- https://bibliotekanauki.pl/articles/38702595.pdf
- Data publikacji:
- 2023
- Wydawca:
- Instytut Podstawowych Problemów Techniki PAN
- Tematy:
-
musculoskeletal image
image processing
image enhancement
fracture diagnosis
fracture classification
deep neural network
obraz układu mięśniowo-szkieletowego
przetwarzanie obrazu
wzmocnienie obrazu
diagnoza złamania
klasyfikacja złamań
głęboka sieć neuronowa - Opis:
- Bone fractures break bone continuity. Impact or stress causes numerous bone fractures. Fracture misdiagnosis is the most frequent mistake in emergency rooms, resulting in treatment delays and permanent impairment. According to the Indian population studies, fractures are becoming more common. In the last three decades, there has been a growth of 480 000, and by 2022, it will surpass 600 000. Classifying X-rays may be challenging, particularly in an emergency room when one must act quickly. Deep learning techniques have recently become more popular for image categorization. Deep neural networks (DNNs) can classify images and solve challenging problems. This research aims to build and evaluate a deep learning system for fracture identification and bone fracture classification (BFC). This work proposes an image-processing system that can identify bone fractures using X-rays. Images from the dataset are pre-processed, enhanced, and extracted. Then, DNN classifiers ResNeXt101, InceptionResNetV2, Xception, and NASNetLarge separate the images into the ones with unfractured and fractured bones (normal, oblique, spiral, comminuted, impacted, transverse, and greenstick). The most accurate model is InceptionResNetV2, with an accuracy of 94.58%.
- Źródło:
-
Computer Assisted Methods in Engineering and Science; 2023, 30, 3; 369-385
2299-3649 - Pojawia się w:
- Computer Assisted Methods in Engineering and Science
- Dostawca treści:
- Biblioteka Nauki
Artykuł