From Pixels to Prognosis: How Artificial Intelligence (AI) Analyzes Medical Images for Childhood Pneumonia Detection

10Aug 2024 by Kiara Ekeigwe

The authors, Taofeeq Oluwatosin Togunwa and Abdulhammed Opeyemi Babatunde, are part of a Mentored Child Health Research Project led by the Duke Center for Policy Impact in Global Health and the University of Ibadan Medical Students Association.

Introduction

Pneumonia is the number one infectious cause of death in childhood, causing about 700,000 deaths a year.1  India bears the highest burden, contributing to nearly 20% of global childhood pneumonia deaths.2 Pneumonia remains one of the leading causes of death among children under 5 years in Sub-Saharan Africa (SSA),3 threatening the achievement of target 3.2 of Sustainable Development Goal 3, the target of reducing child mortality to 25 deaths per 1,000 live births or lower by 2030. The region’s high pneumonia mortality is linked to inadequate healthcare resources, poor health financing, and a high prevalence of pathogens that cause pneumonia.4 Risk factors for childhood pneumonia in SSA include malnutrition, insufficient immunization coverage, and exposure to air pollution.

Standard practice for diagnosing pneumonia is through the combination of clinical signs and symptoms, chest imaging (typically radiography), and microbiological testing. However, childhood pneumonia diagnosis in SSA often relies on clinical acumen alone, with tachypnea (increased respiratory rate) being a key indicator.2, 5 When imaging is used, chest radiographs are the most common imaging method and require very low radiation doses. Challenges in using radiography in the diagnosis of childhood pneumonia include diagnostic error,6 incorrect exposure, and poor positioning.7 In low-resource settings, diagnostic challenges are accompanied by inadequate access to appropriate treatment, especially antibiotics and oxygen.5, 8

There is growing interest in using artificial intelligence (AI) to improve the reach, precision, and efficiency of imaging. In this commentary, we discuss the potential of AI in enhancing detection of childhood pneumonia from medical imaging such as chest radiographs.

 

The Role of AI in Medical Imaging

AI refers to methods and algorithms that mimic human intelligence. It can be classified into strong and weak categories: strong AI aims to solve any problem and emulate human intelligence, while weak AI efficiently performs only specific tasks.9 Weak AI systems improve autonomously (i.e., without human oversight) through progressive learning from acquired information, often using machine learning (ML) techniques that are rooted in computational statistics. ML helps machines learn and make decisions based on training data, allowing for predictions without explicit programming.10 Deep learning (DL) is a form of ML that uses interconnected neural networks inspired by biological neurons, to learn from experience, independently classifying data for specific tasks.11 The reader is encouraged to explore a more comprehensive discussion covering historical concepts of AI, state-of-the-art models, and current AI trends.12

DL is revolutionizing medical imaging, offering significant advancements in diagnostics.13, 14 Human interpretation of medical images, predominantly by radiologists, is hindered by subjectivity, variability, and fatigue.15, 16 The shortage of radiologists in low-resource settings is an additional challenge to achieving high diagnostic accuracy, as it limits the amount of time for image review and can lead to missed findings and delayed reports.17 AI—chiefly DL tools—automates image analysis, aiding in pathology detection, disease quantification, and decision support. AI models hold tremendous potential in medical diagnostics for various conditions such as skin cancer,18 breast cancer,19 retinal diseases,20 and colorectal cancer.21 Emerging research suggests that AI can enhance diagnostic accuracy and workflow efficiency.22, 23

 

Could AI Improve the Diagnosis of Childhood Pneumonia?

The most commonly used AI approach for image processing is to use a tool called convolutional neural networks (CNNs), a specialized type of neural network designed for recognizing patterns in two- or three-dimensional images.24 Although there is growing interest in newer, emerging tools, especially visual transformers models,25 our discussion focuses on the well-researched and widely understood CNNs.

Originally implemented by Fukushima in 1980 and formalized by LeCun and colleagues in 1998, CNNs leverage advanced graphic processing units to handle large data sets efficiently.9  They use a special technique called convolution to analyze images more effectively. This method involves using complex mathematical equations that help the network detect patterns, such as edges and shapes, by looking at small parts of an image.

CNNs show significant promise in the diagnosis of diseases such as tuberculosis, pneumoconiosis, pneumonia, and COVID-19.26 Studies show high diagnostic accuracy, with models such as DenseNet201 and customized VGG16 achieving an accuracy of 0.95% and 0.94%, respectively, in detecting childhood pneumonia.27 AI-powered tools have been developed in Italy,28 Saudi Arabia,29 and the United States 30 to detect pneumonia and other lung diseases from chest radiographs.  However, the feasibility and effectiveness of using AI for detecting childhood pneumonia in SSA has not been fully explored. Mahomed and colleagues conducted a study in South Africa to compare the performance of an AI-based computer-aided diagnostic system with board certified radiologists in detecting pneumonia in chest radiographs of  children.31 The AI performed worse than the radiologists, and the authors concluded that although AI could not replace an expert radiologist, it could supplement human reading in resource-limited settings. There are limited studies describing similar interventions in SSA.

The effective implementation of AI in healthcare systems in SSA faces several challenges, including managing vast amounts of unstructured healthcare data, ensuring model generalizability, and addressing ethical concerns such as data usage, security, and privacy.32  Inadequate information and communication technology infrastructure, low digital skills among healthcare professionals, and limited funding further impede adoption of AI in SSA.32 High-performance computers essential for AI are costly. Furthermore, the absence of emotional intelligence in AI models also poses a challenge in patient-centered care.33

To harness AI’s potential in SSA’s healthcare systems, frameworks must address these challenges. Importantly, integrating AI with these systems should be tailored to specific contexts, focusing on ethical principles such as explainable AI and responsible AI. Explainable AI ensures that the decision-making processes of AI systems are transparent and understandable, helping to identify and correct biases.34 Responsible AI encompasses a broader set of principles, including accountability and transparency, aiming to ensure that AI technologies are developed and used in ways that are ethical and equitable, especially considering the diverse contexts in which they are applied.35

 

Conclusion

Rapidly evolving AI technology presents an opportunity to improve early detection of childhood pneumonia in high-burden, low-resource settings. Large clinical trials are needed to develop low-cost effective AI tools to support diagnosis of childhood pneumonia in these settings.

References

  1. UNICEF Child Health Coverage Database [Internet]. 2023 [cited August 2024]. Available from: https://data.unicef.org/topic/child-health/pneumonia/.
  2. Yadav KK, Awasthi S. Childhood Pneumonia: What’s Unchanged, and What’s New? Indian Journal of Pediatrics. 2023;90(7):693-9.
  3. Iuliano A, Aranda Z, Colbourn T, Agwai IC, Bahiru S, Bakare AA, et al. The burden and risks of pediatric pneumonia in Nigeria: A desk-based review of existing literature and data. Pediatric Pulmonology. 2020;55(S1):S10-S21.
  4. Ojuawo OB, Iroh Tam P-Y. Childhood Pneumonia Diagnostics in Sub-Saharan Africa: A Systematic Review. Journal of Tropical Pediatrics. 2022;68(4).
  5. Shittu F, Agwai IC, Falade AG, Bakare AA, Graham H, Iuliano A, et al. Health system challenges for improved childhood pneumonia case management in Lagos and Jigawa, Nigeria. Pediatric Pulmonology. 2020;55(S1):S78-S90.
  6. Porter P, Brisbane J, Tan J, Bear N, Choveaux J, Della P, et al. Diagnostic Errors Are Common in Acute Pediatric Respiratory Disease: A Prospective, Single-Blinded Multicenter Diagnostic Accuracy Study in Australian Emergency Departments. Frontiers in Pediatrics. 2021;9.
  7. Mahomed N, Manduku V, Hlabangana LT, Gong W, Moore D, Fancourt N, et al. Evaluation of paediatric chest X-ray quality in the Pneumonia Aetiology Research for Child Health (PERCH Study), a prospective international multicentre study in 7 developing countries. Wits Journal of Clinical Medicine. 2023;5(2):85-96.
  8. Marangu D, Zar HJ. Childhood pneumonia in sub-Saharan Africa: still a challenge. J Pan Afr Thorac Soc. 2021;2(1):1-3.
  9. Coppola F, Faggioni L, Gabelloni M, De Vietro F, Mendola V, Cattabriga A, et al. Human, All Too Human? An All-Around Appraisal of the “Artificial Intelligence Revolution” in Medical Imaging. Frontiers in Psychology. 2021;12.
  10. Greener JG, Kandathil SM, Moffat L, Jones DT. A guide to machine learning for biologists. Nature Reviews Molecular Cell Biology. 2022;23(1):40-55.
  11. Kim H. Deep Learning. In: Kim H, editor. Artificial Intelligence for 6G. Cham: Springer International Publishing; 2022. p. 247-303.
  12. Barragán-Montero A, Javaid U, Valdés G, Nguyen D, Desbordes P, Macq B, et al. Artificial intelligence and machine learning for medical imaging: A technology review. Physica Medica. 2021;83:242-56.
  13. Rentiya ZS, Mandal S, Inban P, Vempalli H, Dabbara R, Ali S, et al. Revolutionizing Breast Cancer Detection With Artificial Intelligence (AI) in Radiology and Radiation Oncology: A Systematic Review. Cureus. 2024;16(4):e57619.
  14. Gao C, Wu L, Wu W, Huang Y, Wang X, Sun Z, et al. Deep learning in pulmonary nodule detection and segmentation: a systematic review. European Radiology. 2024.
  15. McGraw M. Radiologist Experience, Fatigue Affect Breast Imaging Interpretation. Oncology Times. 2022;44(6):27.
  16. Alexander R, Waite S, Bruno MA, Krupinski EA, Berlin L, Macknik S, et al. Mandating Limits on Workload, Duty, and Speed in Radiology. Radiology. 2022;304(2):274-82.
  17. Iyawe EP, Idowu BM, Omoleye OJ. Corrigendum: Radiology subspecialisation in Africa: A review of the current status. SA Journal of Radiology. 2021;25(1).
  18. Goyal M, Knackstedt T, Yan S, Hassanpour S. Artificial intelligence-based image classification methods for diagnosis of skin cancer: Challenges and opportunities. Computers in Biology and Medicine. 2020;127:104065.
  19. Zhang J, Wu J, Zhou XS, Shi F, Shen D. Recent advancements in artificial intelligence for breast cancer: Image augmentation, segmentation, diagnosis, and prognosis approaches. Seminars in Cancer Biology. 2023;96:11-25.
  20. Saleh GA, Batouty NM, Haggag S, Elnakib A, Khalifa F, Taher F, et al. The Role of Medical Image Modalities and AI in the Early Detection, Diagnosis and Grading of Retinal Diseases: A Survey. Bioengineering. 2022;9(8):366.
  21. Wang KS, Yu G, Xu C, Meng XH, Zhou J, Zheng C, et al. Accurate diagnosis of colorectal cancer based on histopathology images using artificial intelligence. BMC Medicine. 2021;19(1):76.
  22. Panayides AS, Amini A, Filipovic ND, Sharma A, Tsaftaris SA, Young A, et al. AI in Medical Imaging Informatics: Current Challenges and Future Directions. IEEE Journal of Biomedical and Health Informatics. 2020;24(7):1837-57.
  23. Khalifa M, Albadawy M. AI in diagnostic imaging: Revolutionising accuracy and efficiency. Computer Methods and Programs in Biomedicine Update. 2024;5:100146.
  24. Sarvamangala DR, Kulkarni RV. Convolutional neural networks in medical image understanding: a survey. Evolutionary Intelligence. 2022;15(1):1-22.
  25. Shamshad F, Khan S, Zamir SW, Khan MH, Hayat M, Khan FS, et al. Transformers in medical imaging: A survey. Medical Image Analysis. 2023;88:102802.
  26. Akhter Y, Singh R, Vatsa M. AI-based radiodiagnosis using chest X-rays: A review. Frontiers in Big Data. 2023;6.
  27. Field EL, Tam W, Moore N, McEntee M. Efficacy of Artificial Intelligence in the Categorisation of Paediatric Pneumonia on Chest Radiographs: A Systematic Review. Children. 2023;10(3):576.
  28. Avanzato R, Beritelli F, Lombardo A, Ricci C. Lung-DT: An AI-Powered Digital Twin Framework for Thoracic Health Monitoring and Diagnosis. Sensors. 2024;24(3):958.
  29. Hroub NA, Alsannaa AN, Alowaifeer M, Alfarraj M, Okafor E. Explainable deep learning diagnostic system for prediction of lung disease from medical images. Computers in Biology and Medicine. 2024;170:108012.
  30. Farghaly O, Deshpande P. Texture-Based Classification to Overcome Uncertainty between COVID-19 and Viral Pneumonia Using Machine Learning and Deep Learning Techniques. Diagnostics. 2024;14(10):1017.
  31. Mahomed N, van Ginneken B, Philipsen RHHM, Melendez J, Moore DP, Moodley H, et al. Computer-aided diagnosis for World Health Organization-defined chest radiograph primary-endpoint pneumonia in children. Pediatric Radiology. 2020;50(4):482-91.
  32. Mbunge E, Batani J. Application of deep learning and machine learning models to improve healthcare in sub-Saharan Africa: Emerging opportunities, trends and implications. Telematics and Informatics Reports. 2023;11:100097.
  33. Mohanasundari SK, Kalpana M, Madhusudhan U, Vasanthkumar K, Rani B, Singh R, et al. Can Artificial Intelligence Replace the Unique Nursing Role? Cureus. 2023;15(12).
  34. Loh HW, Ooi CP, Seoni S, Barua PD, Molinari F, Acharya UR. Application of explainable artificial intelligence for healthcare: A systematic review of the last decade (2011–2022). Computer Methods and Programs in Biomedicine. 2022;226:107161.
  35. Okolo CT, Aruleba K, Obaido G. Responsible AI in Africa—Challenges and opportunities. Responsible AI in Africa: Challenges and Opportunities. 2023:35-64.