Enc‐Unet: A novel method for Glioma segmentation

• DOI: 10.1002/ima.22822
• Publication Date: 2022-11-05T09:35:00Z
• Authors: Syed Sajid Hussain; Jainy Sachdeva; Chirag Kamal Ahuja; Abhiav Singh
• Keywords: Dice; Fluid-attenuated inversion recovery; Similarity (geometry); Autoencoder

Abstract The diagnosis' treatment planning, follow‐up and prognostication of Gliomas is significantly enhanced on Magnetic Resonance Imaging. In the present research, deep learning‐based variant convolutional neural network methodology proposed for glioma segmentation where pretrained autoencoder acts as backbone to 3D‐Unet which performs task well image restoration. Further, Unet accepts input combination three non‐native MR images (T2, T1CE, FLAIR) extract maximum superior features segmenting tumor regions. weighted dice loss employed, focusses segregating region into regions interest namely whole with oedema (WT), enhancing (ET), core (TC). optimizer preferred in Adam learning rate initially set , progressively reduced by a cosine decay after 50 epochs. parameters are larger extent (up 9.8 M compared 27 M). experimental results show that model achieved Dice similarity coefficients: 0.77, 0.92, 0.84; sensitivity: 0.90, 0.95, 0.89; specificity: 0.97, 0.99, 0.99; Hausdorff95: 5.74, 4.89, 6.00, including ET, WT, TC. This Glioma method efficient segregation tumors.