An efficient convolutional histogram-oriented gradients and deep convolutional learning approach for accurate classification of bone cancer

In our human body bones are the most significant part, which helps people to move and perform several activities. But, the cancer is caused by producing abnormal cell, which is rapidly spread to the whole parts of the body. Bone cancer is one of the critical types due to its malignancy more than other cancers. The approach involves preprocessing and segmentation of input images to remove noise and resize images, followed by feature extraction using a Convolutional histogram of oriented gradients (ConvHisOrGrad). The ROI extraction helps to accurately identify abnormal parts around the cancerous area. The Extreme Deep Convolutional learning machine (Ex-ConVLM) is used for normal and cancerous bone classification based on the texture properties of bone MRI images. The proposed technique was evaluated using a dataset of 220 bone MRIs for tumor classes classified as necrotic, non-tumor, and viable-tumor. Results showed that the proposed technique outperformed existing techniques with the highest accuracy of 97% for the necrotic tumor class, 98.2% for the non-tumor class, and 98.6% for the viable tumor class. The fine-tuned model shows promising performance in detecting malignancy in bone based on histological images. In summary, the proposed technique utilizes deep learning architectures and ROI extraction for the accurate identification of abnormal parts in bone MRI images, achieving state-of-the-art performance in the detection and categorization of bone cancers.