Skip to main content

3D segmentation of colorectal tumors

A deep learning strategy for the 3D segmentation of colorectal tumors from ultrasound imaging


Colorectal cancer remains a leading cause of cancer-related mortality worldwide, highlighting the need for accurate and efficient diagnostic tools. While Deep Learning has shown promise in medical imaging, its application to transrectal ultrasound for colorectal tumor segmentation remains underexplored. Currently, lesion segmentation is performed manually, relying on clinician expertise and leading to significant variability across treatment centers. To overcome this limitations, we propose a novel strategy that addresses both practical challenges and technical constraints, particularly in scenarios with limited data availability, offering a robust framework for accurate 3D colorectal tumor segmentation from ultrasound imaging.

We evaluate eight state-of-the-art models, including convolutional neural networks and transformer-based architectures, and introduce domain-tailored pre- and post-processing techniques such as data augmentation, patching and ensembling to enhance segmentation performance while reducing computational cost. Leading to an average improvement in term of DICE score of 0.423 absolute points (+107%), compared to baseline models, our findings demonstrate the potential of our proposal to improve the accuracy and reliability of ultrasound-based diagnostics for colorectal cancer, paving the way for clinically viable AI-driven solutions.

we proposed a novel Deep Learning-based strategy for 3D colorectal tumor segmentation in TRUS images. The primary and general contribution of our study was not to offer a fully resolved clinical solution, but rather to demonstrate a practical and effective strategy, more from a methodological perspective than a purely performance-oriented one, to address the challenges of limited data availability, acquisition heterogeneity, and resource-constrained clinical scenarios.

The integration of the various steps of our pipeline was intentionally developed to enhance robustness under such difficult conditions, to aim at enhancing diagnostic and decision support for colorectal cancer diagnosis. Through the proposed combination of techniques, we show that meaningful improvements in segmentation accuracy can be achieved even with very limited training data — an insight that may be valuable not only for 3D TRUS but also for other clinical scenarios facing similar data limitations. Other important contributions are in what follows.

Expanding datasets and refining models will be critical to enhancing clinical applicability. Indeed, a key limiting factor remains the severe scarcity of annotated 3D TRUS data for colorectal cancer segmentation, both in our specific case and more generally across a wide range of complex or rare imaging modalities. Similar challenges are commonly encountered in many other clinical domains characterized by limited data availability, such as rare diseases, expensive or operator-dependent imaging protocols, or underutilized modalities for which expert annotation requires highly specialized personnel.

With continued efforts to overcome these challenges, the development of AI-driven solutions for colorectal cancer diagnosis can lead to more accurate, standardized, and accessible diagnostics, ultimately improving patient care across diverse healthcare environments.

Ultrasound imaging, sonography, diagnostic ultrasound, medical imaging, prenatal scan, fetal ultrasound, Doppler ultrasound, abdominal ultrasound, pelvic ultrasound, transvaginal ultrasound, 3D ultrasound, 4D ultrasound, echocardiogram, ultrasound technician, ultrasound scan, soft tissue imaging, vascular ultrasound, ultrasound gel, real-time imaging, non-invasive scan

#ultrasound, #sonography, #medicalimaging, #fetalultrasound, #pregnancyultrasound, #diagnosticultrasound, #3Dultrasound, #4Dultrasound, #Dopplerultrasound, #echocardiogram, #abdominalultrasound, #pelvicultrasound, #ultrasoundtech, #noninvasive, #ultrasoundscan, #ultrasoundmachine, #imagingtech, #ultrasoundlife, #vascularultrasound, #realtimeimaging

International Conference on Network Science and Graph Analytics

Visit: networkscience.researchw.com

Award Nomination: networkscience-conferences.researchw.com/award-nomination/?ecategory=Awards&rcategory=Awardee

For Enquiries: support@researchw.com


Get Connected Here
---------------------------------
---------------------------------
instagram.com/network_science_awards
tumblr.com/emileyvaruni
n.pinterest.com/network_science_awards
networkscienceawards.blogspot.com
youtube.com/@network_science_awards

Comments

Post a Comment

Popular posts from this blog

Global Lighthouse Network

Smart, sustainable manufacturing: 3 lessons from the Global Lighthouse Network Launched in 2018, when more than 70% of factories struggled to scale digital transformation beyond isolated pilots, the Global Lighthouse Network set out to identify the world’s most advanced production sites and create a shared learning journey to up-level the global manufacturing community. In the past seven years, the network has grown from 16 to 201 industrial sites in more than 30 countries and 35 sectors, including the latest cohort of 13 new sites. This growing community of organizations is setting new standards for operational excellence, leveraging advanced technologies to drive growth, productivity, resilience and environmental sustainability. But what exactly is a Global Lighthouse and what has the network achieved? What is the Global Lighthouse Network? The Global Lighthouse Network is a community of operational facilities and value chains that harness digital technologies at scale to ac...
Post a Comment
Read more

Multi-Modal Data

Multi-Task Federated Split Learning Across Multi-Modal Data with Privacy Preservation With the advancement of federated learning (FL), there is a growing demand for schemes that support multi-task learning on multi-modal data while ensuring robust privacy protection, especially in applications like intelligent connected vehicles. Traditional FL schemes often struggle with the complexities introduced by multi-modal data and diverse task requirements, such as increased communication overhead and computational burdens. In this paper, we propose a novel privacy-preserving scheme for multi-task federated split learning across multi-modal data (MTFSLaMM). Our approach leverages the principles of split learning to partition models between clients and servers, employing a modular design that reduces computational demands on resource-constrained clients. To ensure data privacy, we integrate differential privacy to protect intermediate data and employ homomorphic encryption to safeguard client m...
Post a Comment
Read more

Intelligent visual

Intelligent visual question answering in TCM education: An innovative application of IoT and multimodal fusion This paper proposes an innovative Traditional Chinese Medicine Ancient Text Education Intelligent Visual Question Answering System ( TCM-VQA IoTNet ), which integrates Internet of Things (IoT) technology with multimodal learning to achieve a deep understanding and intelligent question answering of both the images and textual content of traditional Chinese medicine ancient texts. The system utilizes the VisualBERT model for multimodal feature extraction, combined with Gated Recurrent Units (GRU) to process time-series data from IoT sensors, and employs an attention mechanism to optimize feature fusion, dynamically adjusting the question answering strategy. Experimental evaluations on standard datasets such as VQA v2.0, CMRC 2018, and the Chinese Traditional Medicine Dataset demonstrate that TCM-VQA IoTNet achieves accuracy rates of 72.7%, 69.%, and 75.4% respectively, with F1-...
Post a Comment
Read more