LMNS-Net: Lightweight Multiscale Novel Semantic-Net deep learning approach used for automatic pancreas image segmentation in CT scan images

In the study and research of medical images, the sharp and smooth pancreatic segmentation challenge is a critical and challenging one. The most widely utilized and effective technique for pancreatic segmentation with smooth and precise results is a proposed LMNS-net deep learning, bottom-up approach. The proposed LMNS-net is used to automatically segment the pancreas in clinical abdominal computed tomography (CT) images. For the segmentation of the acute pancreas using several angles of CT scans, such as coronal, axial, and sagittal, a proposed LMNS-net model is used. In the LMNS-Net model, 12 layers are used with 4 convolution layers. LMNS-Net model is used for many organ segmentation from CT scans clinical images with high accuracy. The lightweight multiscale block is used in the proposed approach which is aggregating the required feature only so unused information dropout at the convolution layer. The computation time-period is reduced as related to the state-of-art. Validation is 99.78% and loss values vary from 1 to 0 only. The LMNS-Net model achieved a dice similarity index score up to 88.68 & PLUSMN;57.49%. The LMNS-Net model takes 1 to 3 s for the segmentation of medical CT images in the testing process. The LMNS-Net model takes very less time for testing purposes as compared to other approaches. Top-down approaches are not useful in the detection of pancreatic cancer images from CT scan images. In Bottom-Up approaches, the LMNS-Net is useful to detect accurate sharpness of the pancreas and kidney, the pancreatic cancer-affected area within less time only. In medical image segmentation, the time of training deep learning approach is a key constraint. "VGG-16"and "VGG-19"are taking too much time, so more time is required for disease diagnosis. The training time of the LMNS-Net model is reduced as compared to the State-of-Arts.