Figure 1

(a) The complete clustering process (first approach) of segmenting tumors. During this process, the GMM algorithm groups all values within the flattened input array into one of six clusters. The cluster with the highest number of members is then designated as representing healthy pixels, while the other clusters are categorized as suspected pixels. Next, the resulting array is reshaped into a 2D array, and the upper half of this new representation is subjected to post-processing using median and morphology filters. (b) Suggested convolutional neural network for refinement of segmentation results. The proposed neural network architecture comprises three individual subnetworks, each consisting of three parallel convolutional layers. The output of each layer is then combined to form a unified array, which serves as the input for the subsequent subnetwork. This hierarchical structure allows the network to extract increasingly complex features and patterns from the input data, enabling it to learn high-level representations of the segmented images. Moreover, the parallel architecture of the subnetworks helps speed up the computation process while maintaining the accuracy of the predictions.