Elsevier

Clinical Imaging

Volume 80, December 2021, Pages 72-76
Clinical Imaging

Artificial Intelligence
Understanding artificial intelligence based radiology studies: CNN architecture

https://doi.org/10.1016/j.clinimag.2021.06.033Get rights and content

Highlights

  • Artificial intelligence based neural network architectures has yielded high performance in computer vision related tasks.

  • This article explains certain basic concepts of network architecture and its potential role in medical imaging analysis.

  • Tailored CNN architectures have been developed for different tasks including classification, detection and segmentation.

Abstract

Artificial intelligence (AI) in radiology has gained wide interest due to the development of neural network architectures with high performance in computer vision related tasks. As AI based software programs become more integrated into the clinical workflow, radiologists can benefit from better understanding the principles of artificial intelligence. This series aims to explain basic concepts of AI and its applications in medical imaging. In this article, we will review the background of neural network architecture and its application in imaging analysis.

Introduction

Artificial intelligence (AI) is a broad umbrella term used to encompass a wide variety of subfields dedicated to creating algorithms to perform tasks that mimic human intelligence. Deep learning is a type of machine learning inspired by the biological nervous system that uses multiple layers of simple processing units called nodes that are intricately interconnected. Similar to the function of nodes in the nervous system, deep learning algorithms automatically extract and combine progressively higher-level representations of an input in order to perform the task required. There is excitement around deep learning's ability to discover previously unknown relationships in data and perform complex tasks. While deep learning is conceptually not a new idea,1 the application of this technology became a reality in the late 2000's with the advancements in graphics processing units (GPUs), which enabled complex computational requirements of multiple matrix multiplication with parallel processing.

In 2012, another breakthrough came in the form of a new convolutional neural network (CNN) introduced at the annual ImageNet challenge competition. This CNN, named the AlexNet,3 used an innovative architecture to achieve a performance significantly better than previous years. AlexNet provided a new insight into the framework of how neural networks can be developed and became a foundation for many of the currently used networks. Later in 2017, Google released tensor processing units (TPUs) that offered even more computational power.2 This resulted in lower cost and greater accessibility to these hardware developments, enabling further progress in the field of deep learning.

This technology has gained interest in the field of medical imaging and has been applied to tumor detection, pathologic correlation, and treatment response monitoring.[4], [5], [6],7., [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18] Preliminary studies have yielded promising results with performance in some detection tasks comparable to a trained radiologists.19 In this article, we will review general principles of neural network architecture and its relevance towards radiologic image analysis.

Section snippets

Convolutional neural network architecture basics and terms

Neural networks were inspired by the biological nervous system composed of neuronal connections. These networks were further developed mimicking the functionality of the human visual system20,21 which operates by progressively combining the input of neurons that recognize simple features such as lines or colors in the visual field into more complex representations. Fig. 1 illustrates the general foundation of a CNN architecture.

Review of architectures

Many types of CNN architectures have been developed that are tailored for different tasks including classification, detection, and segmentation.

Conclusion

Deep learning technology has made remarkable progress in the last decade and many medical studies have found success using neural networks to perform classification, detection, and segmentation. Understanding basics concepts of neural network architecture and its application in imaging analysis will be essential for radiologists as AI based software programs become more integrated into the clinical workflow.

Declaration of competing interest

No conflict of interest.

References (63)

  • Y.K. Tsehay et al.

    Convolutional neural network based deep-learning architecture for prostate cancer detection on multiparametric magnetic resonance images

    Proc SPIE

    (2017)
  • J. Long et al.

    Fully convolutional networks for semantic segmentation

  • P. Moeskops et al.

    Deep learning for multi-task medical image segmentation in multiple modalities

  • J.Z. Cheng et al.

    Computer-aided diagnosis with deep learning architecture: applications to breast lesions in US images and pulmonary nodules in CT scans

    Sci Rep

    (2016)
  • Y. Ding et al.

    A deep learning model to predict a diagnosis of Alzheimer disease by using 18F-FDG PET of the brain

    Radiology

    (2019)
  • M.A. Mazurowski et al.

    Deep learning in radiology: an overview of the concepts and a survey of the state of the art with focus on MRI

    J Magn Reson Imaging

    (2019)
  • J.W. Patriarche et al.

    Part 1. Automated change detection and characterization in serial MR studies of brain-tumor patients

    J Digit Imaging

    (2007)
  • R. Ha et al.

    Axillary lymph node evaluation utilizing convolutional neural networks using MRI dataset

    J Digit Imaging

    (2018)
  • R. Ha et al.

    Accuracy of distinguishing atypical ductal hyperplasia from ductal carcinoma in situ with convolutional neural network–based machine learning approach using mammographic image data

    Am J Roentgenol

    (2019)
  • R. Ha et al.

    Predicting breast cancer molecular subtype with MRI dataset utilizing convolutional neural network algorithm

    J Digit Imaging

    (2019)
  • D. Ardila et al.

    End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography

    Nat Med

    (2019)
  • K. Fukushima

    Neocognitron: a self organizing neural network model for a mechanism of pattern recognition unaffected by shift in position

    Biol Cybern

    (1980)
  • Y. Lecun et al.

    Gradient-based learning applied to document recognition

    Proc IEEE

    (1998)
  • C. Olah et al.

    Feature visualization

    Distill

    (2017)
  • Jain & Levy et al.

    Breast mass classification using deep convolutional neural networks

  • Jiang

    Jiang F. ICBCB ‘17. Hong Kong: 2017. Breast Mass Lesion Classification in Mammograms by Transfer Learning

    (2017)
  • S. Li et al.

    Predicting lung nodule malignancies by combining deep convolutional neural network and handcrafted features

    Phys Med Biol

    (2019)
  • A. Abdullahi et al.

    Pretrained convolutional neural networks for Cancer genome classification

  • Y. Sun et al.

    Identification of 12 cancer types through genome deep learning

    Sci Rep

    (2019)
  • A. Gertych et al.

    Convolutional neural networks can accurately distinguish four histologic growth patterns of lung adenocarcinoma in digital slides

    Sci Rep

    (2019)
  • M. Lin et al.

    Network in network

  • Cited by (0)

    View full text