Brain diagnosis is a crucial part of medical practice. Several methods exist for making the diagnosis. Brain imaging may be performed using a variety of diagnostic tools, such as CT, MRI, X-ray, and CTA. The diagnostic process is easy if the medical application is quick and realistic. These can save a patient’s life. In this paper, CNN-GB method is proposed to develop a CT-scan based brain diagnosis system. A higher success rate in therapy is possible because to a more thorough and accurate brain diagnostic procedure. Machine and deep learning algorithms detected anomalies in the brain. The patient needs CT scans of his brain analyzed using convolutional neural network and generative model based learning methods to pinpoint the location of the lesion. Accuracy values of 0.992 and 0.993 are achieved using the suggested technique, which are respectable when compared to other approaches. The study will develop a brain diagnostic application based on CT scans. This framework is based on a fusion of CNN deep learning models, gradient boosting machine learning, and Gaussian filters. Combining three algorithms into one produces more precise results than single algorithms or single models. A PSNR of 38.19, an SSIM of 0.83, an accuracy of 0.99, an error rate of 0.07, and a mean arrival time of 0.07 are all attained in this implementation. Moreover, the method is validated at various high noise densities and for all types of densities the implemented application achieves greater improvement. One of the most crucial and difficult jobs in medical imaging is the segmentation of brain tumours. Manual classification by humans can lead to inaccurate predictions and diagnoses. Brain tumours are highly variable in appearance and the similarity of tumour and normal tissue makes it difficult to extract tumour regions from images. Using statistical information derived from the texture, the goal of this study is to identify healthy and pathological pixels very effectively.