Cancer Prediction for Industrial IoT 4.0

A Machine Learning Perspective

Meenu Gupta, Rachna Jain, Arun Solanki, Fadi Al-Turjman, Meenu Gupta, Rachna Jain, Arun Solanki, Fadi Al-Turjman

• Editorial: Chapman and Hall/CRC
• Año: 2021
• ISBN:
  9781000508666
• Edición: 1
• Categoría: Ciencia de la computación
• Categoría: Minería de datos

Cancer Prediction for Industrial IoT 4.0: A Machine Learning Perspective explores various cancers using Artificial Intelligence techniques. It presents the rapid advancement in the existing prediction models by applying Machine Learning techniques. Several applications of Machine Learning in different cancer prediction and treatment options are discussed, including specific ideas, tools and practices most applicable to product/service development and innovation opportunities. The wide variety of topics covered offers readers multiple perspectives on various disciplines.

Features

• Covers the fundamentals, history, reality and challenges of cancer

• Presents concepts and analysis of different cancers in humans

• Discusses Machine Learning-based deep learning and data mining concepts in the prediction of cancer

• Offers real-world examples of cancer prediction

• Reviews strategies and tools used in cancer prediction

• Explores the future prospects in cancer prediction and treatment

Readers will learn the fundamental concepts and analysis of cancer prediction and treatment, including how to apply emerging technologies such as Machine Learning into practice to tackle challenges in domains/fields of cancer with real-world scenarios. Hands-on chapters contributed by academicians and other professionals from reputed organizations provide and describe frameworks, applications, best practices and case studies on emerging cancer treatment and predictions.

This book will be a vital resource to graduate students, data scientists, Machine Learning researchers, medical professionals and analytics managers.

Chapter 1 Investigation of IoMT-Based Cancer Detection and Prediction

Meet Shah, Harsh Patel, Jai Prakash Verma, and Rachna Jain

CONTENTS

1. 1.1 Introduction
2. 1.2 Cancer Diagnosis and Research
3. 1.2.1 Computational Analysis for Cancer Research
4. 1.2.2 Role of the IoMT in Cancer Detection and Prediction
5. 1.2.3 Role of ML/DL Techniques in Cancer Detection and Prediction
6. 1.3 Literature Review
7. 1.4 Proposed Methodology
8. 1.5 Transfer Learning
9. 1.5.1 Pre-Trained Models
10. 1.5.2 VGG16 and VGG19
11. 1.5.3 ResNet-50
12. 1.5.4 DenseNet-121
13. 1.6 Experiment Setting
14. 1.6.1 Source of Dataset
15. 1.6.2 Feature Extraction and Classification
16. 1.6.3 Pre-Processing and Training
17. 1.6.4 Model Evaluation Metrics
18. 1.7 Results and Comparative Analysis
19. 1.8 Summary
20. References

DOI: 10.1201/9781003185604-1

1.1 Introduction

According to the World Health Organization, the top two causes of cancer death in 2020 were lung cancer and colon and rectum cancer [1]. A quick, safe, and accurate early-stage cancer diagnosis can save millions of lives through proper diagnosis and treatment. There are many screening procedures for cancer diagnosis that work under different conditions and for different types of cancer. Medical practitioners analyze these results for patient diagnosis and treatment, but a misinterpretation of the data or human error can lead to misdiagnosis. To combat these issues, machine learning and deep learning (ML/DL) techniques, combined with the Internet of Medical Things (IoMT), have been a boon for cancer prediction, diagnosis, treatment, and patient care. The majority of this research has been carried out on breast, lung, and prostate cancer [2]. DL architectures generally suffer from tuning a huge number of parameters and require huge data to train. The goal of this chapter is to propose an IoMT-driven DL framework for detection and classification of the top two deadliest cancers – lung cancer and colon cancer – using transfer learning. A publicly available dataset, i.e., LC25000 [3], is used for this research. Upon feature extraction from pre-trained convolutional neural network (CNN) models like VGGNet [4], ResNet50 [5], and DenseNet121 [6], features were fed into a dense and flattened layer for cancer and its sub-type classification depending on the type of cancer. We evaluated the proposed approach by analyzing accuracy, precision, recall, and F1 score. The results showed that CNN pre-trained model ResNet-50 achieved the highest classification rate of 98.53% and 99.93% for the lung cancer and colon cancer datasets, respectively.

1.2 Cancer Diagnosis and Research

Cancer is a genetic disease caused by the unregulated growth of normal cells into tumor cells that happens in a multistage process. According to the World Health Organization, almost 19.3 million cancer cases were reported in 2020, and an estimated 28.4 million cases are projected to occur in 2040 [7]. Cancer is more likely to be treated successfully if diagnosed at an early stage. Moreover, early detection and diagnosis also decrease mental and physical pain suffered by patients. For example, 9 out of 10 patients diagnosed with lung cancer at its earliest stage survive for at least 1 year, which is reduced to just 2 out of 10 patients when lung cancer is diagnosed at the most advanced stage [8]. To detect lung cancer, there are many non-invasive imaging techniques such as computed tomography (CT) scan, chest magnetic resonance imaging (MRI) scan, and positron emission tomography (PET) scan. To detect colon cancer, there is CT scan, colonography, and PET scan. Although these imaging methods can show the tumor size, shape, and position, they are sometimes followed by a biopsy to further determine whether the tumor is cancerous and the grade of cancer. Types of biopsy include endoscopic biopsy, where the doctor inserts a thin, flexible tube called an endoscope via the patient’s mouth or rectum to look for the tumor and also to collect a small tissue sample; needle biopsy, where the doctor uses a special needle to collect tissue samples from a suspicious area; or surgical biopsy, where the doctor makes an incision in the skin to access and collect tissue samples from a suspicious area. These biopsies provide a histological assessment of the microscopic structure of the tissue, and pathologists make the final diagnosis based on a visual inspection of histological samples under a microscope [9]. The process of microscopic examination of tissue to diagnose a disease is called histopathology [10].

1.2.1 Computational Analysis for Cancer Research

The rise in various forms of cancers and other illnesses has made pathologists key supporters in the medical industry, and doctors rely on them for accurate and efficient diagnosis. But the histopathological analysis is time-consuming, requires experience, and is prone to human error when done by pathologists [11]. Therefore, we propose computer-aide...