KGiSL Institute of Technology

(Approved by AICTE, New Delhi; Affiliated to Anna University, Chennai)

Recognized by UGC, Accredited by NBA (IT)
365, KGiSL Campus, Thudiyalur Road, Saravanampatti, Coimbatore – 641035.

LIVER TUMOR SEGMENTATION USING

DEEP LEARNING

711720205002 - AKASH S
711720205008 - DANIEL A
711720205009 - DANUSH S
711720205310 – SANJAI PRASATH S

GUIDED BY
Ms.SHIRLEY JOSEPHINE MARY R M.E.(Ph.D.)
Assistant Professor/IT
Abstract
• At its essence, our liver tumor segmentation model strives to revolutionize
medical imaging by enabling accurate identification and delineation of tumors
within the liver.

• By enabling accurate and automated segmentation of liver tumors from medical

scans, our platform enhances clinical efficiency and minimizes human error,
leading to early detection and improved patient outcomes.
Objective
• The objective of employing deep learning for liver tumor segmentation is to
accurately and efficiently delineate tumor boundaries from medical images, such
as CT and MRI scans.
• By automating the segmentation process, deep learning can reduce human error
and streamline clinical workflows. The goal is to facilitate early detection and
precise treatment planning for liver cancer, ultimately improving patient
outcomes.
Problem Statement
• Liver tumor segmentation is a crucial step in diagnosing and planning treatment
for liver cancer, yet traditional methods are manual, time-consuming, and prone
to human error. This leads to inconsistent segmentation results, potentially
impacting patient outcomes.
• The challenge is to develop a deep learning approach that automates liver tumor
segmentation with high accuracy and reliability. Addressing this problem could
streamline clinical workflows and improve early detection and treatment
planning.
Introduction
• Liver tumor segmentation is pivotal for effective diagnosis and treatment
planning in liver cancer management.
• Traditional segmentation methods are labor-intensive and error-prone,
highlighting the need for automated solutions.
• Deep learning techniques offer promising avenues for accurate and efficient
liver tumor segmentation, revolutionizing medical imaging practices.
Existing System
• Convolutional Neural Networks (CNNs), especially U-Net and its variants, are
widely used for liver tumor segmentation with deep learning.These systems are
trained on large datasets of annotated CT and MRI scans to accurately identify
and segment liver tumors.

• Transfer learning, involving pre-trained models fine-tuned with liver-specific

data, enhances system performance and accelerates training. Deep learning-
based segmentation systems are designed to support medical professionals in
diagnosing liver tumors and planning treatment, aiming to improve accuracy and
efficiency while reducing manual effort.
Proposed System
• The proposed system uses DenseNet architecture, which promotes feature reuse
through densely connected layers. In DenseNet, each layer gets inputs from all
preceding layers, improving gradient flow and reducing the risk of vanishing
gradients. This design allows for deeper networks with fewer parameters,
resulting in more efficient learning.

• The proposed system aims for higher accuracy by prioritizing high-quality data,
advanced architectures, and thorough training techniques. Techniques like
transfer learning and multi-modal imaging can boost the model's accuracy and
generalization even further.
Literature Survey
Name of the Paper Author About the work Year
Liver Tumor Segmentation and Classification: Munipraveena Rela, 1. Hepatocellular Carcinoma is the most common
A Systematic Review Nagaraja Rao Suryakari liver cancer, causing most liver cancer-related 2020
deaths. The liver has a complex structure, making
detection challenging.
2. CT is the primary imaging technique for
identifying and treating liver tumors. Early
detection is critical to prevent severe
complications.
3. Accurate liver tumor segmentation from CT
images is vital for diagnosis, pathology, and
anatomical studies..

Deep learning techniques for liver and liver Sidra Gul a b, Muhammad 1. Liver tumor segmentation in 3D volumetric
tumor segmentation Salman Khan images is complex due to similar properties
2022
among neighboring organs like the heart, spleen,
stomach, and kidneys, requiring advanced
techniques for accurate segmentation.
2. Automatic and semi-automatic methods for liver
segmentation have emerged, with deep learning
playing a crucial role in enhancing diagnosis and
detection of liver tumors.
3. This study reviews deep learning approaches for
segmenting the liver and detecting liver tumors,
focusing on evaluation metrics like accuracy and
dice coefficient.
7
 Name of the Paper Author About the work Year
A Deep Learning Approach for Liver and Hameedur Rahman ,Tanvir 1. Liver cancer is the ninth most common cancer in
Tumor Segmentation in CT Images Using Fatima Naik Bukht women, with liver and liver tumor segmentation 2020
ResUNet being challenging due to overlapping intensities
and varying tissue shapes. Traditional
segmentation methods are often inadequate for
CT images.
2. This study introduces a hybrid ResUNet model,
combining ResNet and UNet architectures, to
improve liver tumor segmentation from CT image
volumes. This model addresses the limitations of
traditional methods, focusing on accurate region
of interest (ROI) assessment.

deep learning methods for liver segmentation Kumar S. S., Vinod Kumar 1. This literature survey explores recent
from CT images R. S. developments in deep learning for segmenting the
2023
liver from CT images, focusing on various
architectures like CNNs, U-Net, attention
mechanisms, GANs, and transformer models,
while also addressing evaluation metrics and
common benchmark datasets.
2. The survey discusses challenges such as
interpretability, model robustness, and ethical
considerations, concluding with a summary of
key findings and suggestions for future research,
aiming to bridge the gap between deep learning
advancements and clinical applications in liver
segmentation.
7
Modules
• Interface Module
• Data Preprocessing Module
• Feature Extraction and Representation Module
• Dashboard
Model Training Module
• Model Evaluation Module
• Deployment and Integration Module
Technical Requirements
Software Requirements Description/Specification/Model Version

 Operating System  Windows

 Frontend Technologies  HTML, CSS

 Database  MySql

 Backend Technologies  Python , Flask framework, Deep

learning architecture
Architecture Diagram
Use Case Diagram
Class Diagram
Front End Designs
Front End Designs - Register
Front End Designs – Login
Front End Designs - Dashboard
Database Designs
Experimental Results
Conclusion

• Deep learning has demonstrated potential for improving the accuracy and
efficiency of liver tumor segmentation, offering automated solutions that reduce
human error and streamline clinical workflows.

• To further advance these benefits, ongoing research should focus on enhancing

generalization and reducing computational complexity, ensuring these models
can be effectively integrated into clinical practice to support better patient
outcomes.
Future Enhancement
• Future enhancements for liver tumor segmentation using deep learning could
focus on leveraging transfer learning to reduce training time and improve model
generalization. Incorporating multi-modal imaging, combining CT and MRI
data, can also enhance segmentation accuracy by providing a more
comprehensive view.

• Another area for improvement is interpretability, with techniques like attention

mechanisms making deep learning models more understandable to clinicians.
Addressing computational complexity through methods like pruning and
quantization can lead to more efficient models, facilitating smoother integration
into clinical workflows.
Reference
1. Christ, P. F., Elshaer, M. E. A., Ettlinger, F., Tatavarty, S., Bickel, M., Bilic,
P., ... & Ahmadi, S. A. (2016). Automatic liver and lesion segmentation in CT
using cascaded fully convolutional neural networks and 3D conditional random
fields. In International Conference on Medical Image Computing and
Computer-Assisted Intervention (pp. 415-423). Springer, Cham.
2. Ben-Cohen, A., Klang, E., Amitai, M. M., & Greenspan, H. (2017). Fully
convolutional network for liver segmentation and lesions detection. In Deep
Learning and Data Labeling for Medical Applications (pp. 77-85). Springer,
Cham.
3. Vorontsov, E., Cerny, M., Urschler, M., Tang, Y. B., & Abi-Jaoudeh, N. (2018).
LiTS—Liver Tumor Segmentation in CT scan images using convolutional
neural networks. In International MICCAI Workshop on Computational and
Clinical Applications in Deep Learning (pp. 50-60). Springer, Cham.
THANK YOU