Brain Tumor Segmentation Using U-Net 3+

Abstract

Themeasurementoftumourextentisadifficulttaskinbraintumourtreatmentplanning and quantitative evaluation. Non-invasive magnetic resonance imaging (MRI) has evolved as a first-line diagnostic method for brain malignancies that does not require ionising radiation. The manualsegmentationofbraintumour extent from3DMRI volumes isatime- consuming jobthatheavilyreliesontheoperator\'sknowledge.Inthiscontext,adependablefullyautomatic segmentation approach for brain tumour segmentation is required for accurate tumour extent determination. Inthisworkweoffer a fullyautomatic method for braintumour segmentation, whichis basedonU-Net-baseddeepconvolutionalneuralnetworks.Ourtechnique wastested using the Multimodal Brain Tumor Image Segmentation (BRATS 2018) datasets, which included 220 cases of high-grade brain tumour and 54 cases of low-grade tumour. Cross- validation has demonstrated that our method efficiently obtains promising segmentation.

Introduction

The text discusses primary malignant brain tumors, focusing on gliomas, which make up about 80% of these tumors and vary from low-grade to highly aggressive types like glioblastoma. Despite advances in medical imaging, surgery, chemotherapy, and radiotherapy, high-grade tumors still have poor survival rates, with glioblastoma showing less than 10% survival at 2.5 years.

Brain tumors are mainly categorized into primary central nervous system lymphomas and gliomas. MRI plays a crucial role in tumor diagnosis and treatment planning, though challenges remain in accurately segmenting tumors due to their variability in size, shape, and texture. Advanced image segmentation is key for precise tumor delineation and treatment.

The text also reviews related research on glioma epidemiology, survival rates, and tumor cell infiltration. It then introduces a methodology for brain tumor image segmentation using the U-Net 3+ deep learning architecture, which enhances segmentation accuracy by combining encoder-decoder paths with dense skip connections.

The U-Net 3+ architecture includes contracting (encoding), bridge, and expansive (decoding) paths to capture local and global image features. Preprocessing steps like normalization and contrast enhancement improve input quality. The architecture uses multi-scale feature fusion and deep supervision for refined segmentation output.

The BraTS 2018 dataset, containing multiple MRI sequences (T1, T2, T1-enhanced, FLAIR) and annotated tumor masks, is used for training and validating the model.

Conclusion

Inconclusion,thisprojectfocusedonthesegmentationofbraintumorimagesusingtheadvanced U-Net 3+ architecture. We successfully completed the task by employing a dataset specifically curated for brain tumor segmentation. The U-Net 3+ architecture, known for its abilityto capture detailed informationat different scales, was trained using the dataset(BRATS 2018). During the training phase, the modellearned to accuratelyidentifyandsegmenttumorregionsinthebrainimages.Thissegmentationprocessinvolved differentiating the veins from the tumor regions, effectively highlighting the areas of interest.

References

[1] Schwartzbaum, J.A., Fisher, J.L., Aldape, K.D., Wrensch, M.: Epidemiologyand molecular pathologyof glioma. Nat. Clin. Pract. Neurol. 2, 494–503 (2006) [2] Smoll, N.R., Schaller, K., Gautschi, O.P.: Long-term survival of patients with glioblastoma multiforme (GBM). J. Clin. Neurosci. 20, 670–675 (2013) [3] Ramakrishna, R., Hebb, A., Barber, J., Rostomily, R., Silbergeld, D.: Outcomes in reoperated low-grade gliomas. Neurosurgery 77, 175–184 (2015) [4] Mazzara,G.P.,Velthuizen,R.P.,Pearlman,J.L.,Greenberg,H.M.,Wagner,H.:Braintumortargetvolume determinationforradiationtreatmentplanningthroughautomatedMRIsegmentation.Int.J.Radiat.Oncol. Biol. Phys. 59, 300–312 (2004) [5] Yamahara,T.,Numa,Y.,Oishi,T.,Kawaguchi,T.,Seno,T.,Asai,A.,Kawamoto,K.:Morphologicaland flow cytometric analysis of cell infiltration in glioblastoma: a comparison of autopsy brain and neuroimaging. BrainTumor Pathol. 27,81–87(2010) [6] Bauer, S., Wiest, R., Nolte, L.-P., Reyes, M.: A survey of MRI-based medical image analysis for brain tumorstudies.Phys.Med.Biol.58,R97–R129(2013)AutomaticBrainTumorDetectionandSegmentation 515 [7] Jones,T.L.,Byrnes,T.J.,Yang,G.,Howe,F.A.,Bell,B.A.,Barrick,T.R.:Braintumorclassificationusing the diffusion tensor image segmentation (D-SEG) technique. Neuro. Oncol. 17, 466–476 (2014) [8] Soltaninejad,M.,Yang,G.,Lambrou,T.,Allinson,N.,Jones,T.L.,Barrick,T.R.,Howe, F. A., Ye, X.: Automated brain tumour detection and segmentation using superpixel- based extremely randomized trees in FLAIR MRI. Int. J. Comput. Assist. Radiol. Surg. 12(2), 183– 203 (2016) [9] Szilágyi, L.,Lefkovits,L.,Benyó,B.:Automatic braintumorsegmentationin multispectralMRI volumes using a fuzzy c-means cascade algorithm. In: 2015 12th International Conference on Fuzzy Systems and Knowledge Discovery (FSKD), pp. 285– 291 (2015) [10] Mei, P.A., de Carvalho Carneiro, C., Fraser, S.J., Min, L.L., Reis, F.: Analysis of neoplastic lesions in magnetic resonance imaging using self-organizing maps. J. Neurol. Sci. 359, 78– 83 (2015)

Copyright

Copyright © 2025 Ch. Meghana. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.