Sim-MedNet: A Simple Deep Learning Framework for the Classification and Comparative Analysis of Medical Image

In a wide range of tests and treatments, X-ray imaging examinations are acknowledged as helpful medical tools. Without having to create an incision, it can assist medical professionals in visualizing the body. They can use this to identify, track, and treat a variety of medical disorders. Deep learning technology has recently been employed in a variety of sectors to analyze medical images, and it excels at tasks like segmentation and registration. In this paper, we proposed a new deep learning framework Sim-MedNet to classify seven human bone images e.g., elbow, finger, forearm, hand, humerus, shoulder, and wrist from the MURA dataset, and compared the result of our proposed model with six other neural network models. We also tried to achieve a decent model performance (maximization of accuracy and minimization of loss) utilizing optimizers like Adam, AdaDelta, AdaGrad, Nadam, RMSprop, SGD Momentum, and SGD Nesterov Momentum. The network architecture was trained to achieve the best possible model parameters, improving training efficiency and exceeding performance levels in the classification tasks of musculoskeletal radiographs. Some state-of-the-art models, like GoogleNet and AlexNet, require significant computational resources and time, making them impractical for real-time or resource-constrained applications. This work addressed the issue and achieved 92.27 % accuracy using the entire MURA dataset and 88.74% accuracy with half of it using our proposed framework, surpassing most state-of-the-art methods. Overall, this work presents a practical, reliable, and efficient approach for medical image diagnosis, offering high accuracy in assisting radiologists.