Bone Fractures Classification Using Deep Learning Approaches

Abstract

Fracture detection in medical imaging is a well-established paradigm. These days, a lot of doctors and other medical professionals utilize computer-aided diagnostic systems (CAD) to assist them diagnose a variety of illnesses more correctly by evaluating medical images. Similarly, typical explanations of bone fractures include trauma, pressure, and osteoarthritis. In addition, bone is a hard substance that supports the entire body. Thus, the bone fracture is regarded as the major problem of the last year. In CAD systems, computerized vision-based bone fracture recognition is growing more and more important since it reduces physician workload by identifying instances that are easy to address. This paper introduces many image processing techniques to identify different forms of fractures in the lower leg bones, the femur and tibia. The purpose of the research is to use an x-ray image to detect the kind of fracture and ascertain if the tibia and femur are both broken. Various approaches and algorithms have been created to accurately detect and classify images based on the presence or absence of fractures in different body parts. In this particular experiment, two distinct class types—Fracture and Normal—as in addition to deep learning-based models were employed. The five models: MobileNetV2, InceptionV3, VGG16, VGG19, and InceptionResNetV2 are utilized to anticipate and identify X-ray pictures in order to classify bone fractures. Finally, the technique's results are assessed using two different performance assessments. Performance evaluation for fractures and normal circumstances is the initial accuracy set, and it uses four possible outcomes: TP, TN, FP, and FN. Using these models, the accuracy of each kind of fracture in mistake scenarios is analyzed next. My suggested method opens the door for autonomous recognition of femur & tibia fractures in bones thanks to the InceptionResNetV2 approach, which has a 94.23% accuracy rate. In the end, the InceptionResNetV2 network is employed for classification in order to recognize fracture in order to generate a web prototype.