Synopsis

Pancreatic cancer is one of the deadliest forms of cancer due to its late diagnosis and aggressive nature. Accurate segmentation of the pancreas from Computed Tomography (CT) scans is a critical step in diagnosing pancreatic cancer, assessing tumor size, and planning treatment strategies. However, manual segmentation of the pancreas is challenging due to its small size, irregular shape, and anatomical variability across patients. In this project, we aim to develop a deep learning-based approach for automated pancreas segmentation in CT images. The model will utilize advanced architectures like 3D U-Net or attention-based CNNs to accurately delineate the pancreas region, enabling early detection of abnormalities and supporting clinical decision-making.

Relevance of the Topic

The pancreas is a difficult organ to segment due to its complex morphology and low contrast in CT images. Automated segmentation can significantly reduce the time and effort required by radiologists, while also improving diagnostic accuracy. Early and precise detection of pancreatic abnormalities, such as tumors, can lead to better patient outcomes through timely intervention. With advancements in deep learning and the availability of large-scale medical imaging datasets, there is an opportunity to create robust models that can assist in the early detection of pancreatic cancer.

Future Research/Scope

• Multi-organ Segmentation: Extend the model to segment other abdominal organs alongside the pancreas, providing a more comprehensive analysis of CT scans.
• Tumor Subtype Classification: Incorporate classification layers to differentiate between benign and malignant pancreatic lesions.
• Cross-modality Integration: Combine CT data with other imaging modalities (e.g., MRI or PET) to improve segmentation and diagnostic accuracy.
• Explainability: Develop techniques to explain model predictions, ensuring transparency and trust in clinical settings.
• Clinical Deployment: Optimize the model for real-time inference to integrate it into clinical workflows.

Skills Learned

• Deep Learning: Experience with 3D U-Net, attention mechanisms, and CNNs for medical image segmentation.
• Medical Image Processing: Knowledge of CT image preprocessing, including intensity normalization, resampling, and augmentation.
• Python Programming: Proficiency in Python and libraries like TensorFlow, PyTorch, and Numpy.
• Data Visualization: Skills in visualizing CT scan slices and segmentation masks using tools like ITK-SNAP or Matplotlib.

Relevant courses to the topic

• CSE428: Image Processing
• CSE422: Artificial Intelligence
• CSE425: Neural Networks
• CSE427: Machine Learning
• CSE443: Bioinformatics I
• Coursera: AI for Medical Diagnosis

Reading List/Study Materials

• Books
  • "Deep Learning for Medical Image Analysis" – Zhou, Greenspan, Shen (Link)
  • "Deep Learning for Biomedical Data Analysis" – Mourad Elloumi (Link)
• Research Papers
  • "Pancreatic cancer detection through semantic segmentation of CT images: a short review" – Karri et al., Discover Artificial Intelligence
    Link
  • "Improved Pancreatic Cancer Detection and Localization on CT Scans: A Computer-Aided Detection Model Utilizing Secondary Features" – Ramaekers et al., Cancers
    Link
  • "Pancreatic Cancer Detection on CT Scans with Deep Learning: A Nationwide Population-based Study" – Chen et al., Radiology
    Link
  • "Semantic segmentation of pancreatic medical images by using convolutional neural network" – Huang et al., Biomedical Signal Processing and Control
    Link
  • "Segmentation of pancreatic ductal adenocarcinoma (PDAC) and surrounding vessels in CT images using deep convolutional neural networks and texture descriptors" – Mahmoudi et al., Nature Scientific Reports
    Link
  • "A Review of Deep-Learning-Based Medical Image Segmentation Methods" – Liu et al., Sustainability
    Link
  • "U-Net: Convolutional Networks for Biomedical Image Segmentation" – Ronneberger et al., MICCAI
    Link
  • "3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation" – Özgün Çiçek et al., MICCAI
    Link
  • "nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation" – Isensee et al., Nature Methods
    Link
  • "Attention U-Net: Learning Where to Look for the Pancreas" – Oktay et al., MIDL
    Link
  • "Image Segmentation Using Deep Learning: A Survey" – Minaee et al., IEEE Transactions on Pattern Analysis and Machine Intelligence
    Link
  • "Deep learning for medical image segmentation: State-of-the-art advancements and challenges" – Rayed et al., Informatics in Medicine Unlocked
    Link
  • "UnSegGNet: Unsupervised Image Segmentation using Graph Neural Networks" – Reddy et al., ArXiv
    Link | Code
  • "Review of Graph Neural Networks for Medical Image" – Jing Wang, EAI Endorsed Transactions on e-Learning
    Link
• Datasets
  • The Medical Segmentation Decathlon (MSD) - Pancreas Dataset (Task 07)
    Link
  • TCIA (The Cancer Imaging Archive) - Pancreas-CT Dataset
    Link
  • Synapse Multi-Organ Segmentation Dataset (MICCAI 2015 Multi-Atlas Abdomen CT)
    Link
• Code Tutorials & Repositories
  • Tutorial: How to Segment a Pancreas CT
  • Code Tutorial: U-Net Implementation in TensorFlow/Keras
• Videos & Playlists
  • Playlist: Medical Imaging with Deep Learning
  • Blog: "Deep Learning with MRI and CT Images" – Towards Data Science