Semantic-Based Optimization of Deep Learning for Efficient Real-Time Medical Image Segmentation

Semantic-Based Optimization of Deep Learning for Efficient Real-Time Medical Image Segmentation

Zhenkun Wei, Jia Liu, Yu Yao
Copyright: © 2024 |Pages: 16
DOI: 10.4018/IJSWIS.340938
Article PDF Download
Open access articles are freely available for download

Abstract

In response to the critical need for advanced solutions in medical imaging segmentation, particularly for real-time applications in diagnostics and treatment planning, this study introduces SM-UNet. This novel deep learning architecture efficiently addresses the challenge of real-time, accurate medical image segmentation by integrating convolutional neural network (CNN) with multilayer perceptron (MLP). The architecture uniquely combines an initial convolutional encoder for detailed feature extraction, MLP module for capturing long-range dependencies, and a decoder that merges global features with high-resolution CNN map. Further optimization is achieved through a tokenization approach, significantly reducing computational demands. Its superior performance is confirmed by evaluations on standard datasets, showing interaction times drastically lower than comparable networks—between 1/6 to 1/10, and 1/25 compared to SOTA models. These advancements underscore SM-UNet's potential as a groundbreaking tool for facilitating real-time, precise medical diagnostics and treatment strategies.
Article Preview
Top

Materials And Methods

Datasets

To evaluate the performance and generalizability of proposed method, experiments were conducted on the widely used Medical Segmentation Decathlon (MSD) dataset (Simpson et al., 2019) and a private rectal cancer magnetic resonance imaging (MRI) dataset (informed consent was obtained from all participants). In addition, to cover the ultrasound database, SM-UNet was also tested on DDTI dataset (Pedraza et al., 2015). To protect participant privacy and confidentiality, any identifying features were cropped out of all images, and no patient details or identifiers have been included in any scans or photographs presented in this paper. The authors confirm that the use of these datasets was in compliance with ethical guidelines and patient privacy and confidentiality were protected.

DDTI Dataset

The DDTI dataset (Pedraza et al., 2015) consists of a set of brightness-mode ultrasound images of the thyroid, including a complete annotation and diagnostic description of suspicious thyroid lesions by expert radiologists. While these lesions include thyroiditis, cystic nodules, adenomas, and thyroid cancer, we only include images with thyroid nodules for our experiments. After preprocessing to remove irrelevant regions and data cleaning using Wang’s method (Wang, 2022), this study collected 637 images along with the corresponding thyroid nodule segmentation maps which were resized to a resolution of 512 ∗ 512.

Medical Segmentation Decathlon

The MSD dataset consists of 10 segmentation tasks from various organs and imaging modalities. These tasks are designed to simulate situations often encountered in medical im- ages, such as small training sets, unbalanced classes, multi-modality data, and small objects (Simpson et al., 2019). SM-UNet and other CNN-based and ViT-based approaches were trained and evaluated on two of these tasks: Task02_Heart (Left atrium segmentation) and Task10_Colon (Colon Cancer segmentation). The modality and resolution of each task after preprocessing and data cleaning can be found in Table 1.

Table 1.
Preprocess for MSD
TaskModalityResolution
Task02_HeartMRI320*320
Task10_ColonCT512*512

Rectal Cancer Dataset

Our private dataset consists of 101 magnetic resonance images (MRI) from patients diagnosed with rectal cancer with corresponding target regions of interest (ROIs) annotated by board-certified radiologists. The ROIs include mesocolic lymph node, mesocolon, and tumor region. This study only used images of tumor region, and only native T2-weighted (T2w) was used. This study collected 1249 images with a resolution of 512 * 512 after preprocessing and data cleaning.

Complete Article List

Search this Journal:
Reset
Volume 20: 1 Issue (2024)
Volume 19: 1 Issue (2023)
Volume 18: 4 Issues (2022): 2 Released, 2 Forthcoming
Volume 17: 4 Issues (2021)
Volume 16: 4 Issues (2020)
Volume 15: 4 Issues (2019)
Volume 14: 4 Issues (2018)
Volume 13: 4 Issues (2017)
Volume 12: 4 Issues (2016)
Volume 11: 4 Issues (2015)
Volume 10: 4 Issues (2014)
Volume 9: 4 Issues (2013)
Volume 8: 4 Issues (2012)
Volume 7: 4 Issues (2011)
Volume 6: 4 Issues (2010)
Volume 5: 4 Issues (2009)
Volume 4: 4 Issues (2008)
Volume 3: 4 Issues (2007)
Volume 2: 4 Issues (2006)
Volume 1: 4 Issues (2005)
View Complete Journal Contents Listing