Improved colorization and classification of intracranial tumor expanse in MRI images via hybrid scheme of Pix2Pix-cGANs and NASNet-large

Mavra Mehmood, Nasser Alshammari, Saad Awadh Alanazi, Asma Basharat, Fahad Ahmad, Muhammad Sajjad, Kashaf Junaid

Research output: Contribution to journalArticlepeer-review


Clinical image processing plays a significant role in healthcare systems and is a widely used methodology of the current era. The Intracranial tumor affects children and adults as it is the 10th most common form of tumor. Accurate tumor segmentation in the brain is complicated even though Intracranial tumor images are acquired properly. The tumor can be curable if its stage and form can be identified on time, and for this purpose, researchers have been developing sophisticated techniques and methods. An automatic detection, segmentation, colorization, and classification of tumor region are done to identify abnormalities in a medical image using T1-CE, an MRI dataset that assists in diagnosis. To assist the medical practitioners in visualizing tumor shape, size, and orientation; only the tumor region is colored in a greyscale image. Pix2Pix Conditional Generative Adversarial Neural Networks (Pix2Pix-cGANs) have generated MRI images with color tumor regions. In qualitative measures, we achieved the Structure Similarity Index (SSIM) average score of 0.92% and an average 28% Peak Signal to Noise Ratio (PSNR) value on generated images that outperform the existing techniques. Moreover, we achieved Classification Accuracy (CA) 88.5% and 92.4% quantitatively in pre and post colorization phases, respectively, with other measures using NASNet-Large.
Original languageEnglish
Pages (from-to)4358-4374
Number of pages17
JournalJournal of King Saud University-Computer and Information Sciences
Issue number7
Early online date28 Jun 2022
Publication statusPublished - 1 Jul 2022


  • Medical imaging
  • Intracranial tumor
  • Magnetic resonance imaging
  • Detection
  • Segmentation
  • Colorization
  • Conditional Generative Adversarial Neural
  • Network
  • Deep learning
  • VGG-16
  • VGG-19
  • NASNet-Large

Cite this