A Special Interview with Prof. Yudong Zhang-"Clarivate 2023 Highly Cited Researcher"
On January 30, 2024, the Editorial Office of the Journal of Cancer Metastasis and Treatment (JCMT) had the privilege of conducting a special interview with one of the Editorial Board members Prof. Yudong Zhang, the Chair Professor at the School of Computing and Mathematical Sciences, University of Leicester, UK. He is the recipient of the "Clarivate 2023 Highly Cited Researchers" award.
Introduction of the Interviewee:
Prof. Yudong Zhang serves as a Chair Professor at the School of Computing and Mathematical Sciences, University of Leicester, UK. His research interests include deep learning and medical image analysis. He holds Fellowships with IET, EAI, and BCS, and is a Senior Member of IEEE, IES, and ACM. He is also a Distinguished Speaker of ACM. He was recognized as Most Cited Chinese Researchers (Computer Science) by Elsevier from 2014 to 2018 and included in the World’s Top 2% Scientist by Stanford University from 2020 to 2022. Additionally, he is a Clarivate Highly Cited Researcher in 2019, 2021, 2022, and 2023. His accolades include the Emerald Citation of Excellence 2017, MDPI Top 10 Most Cited Papers 2015, Information Fusion 2022 Best Paper Award, etc. With over 300 peer-reviewed articles, including more than 50 ESI Highly Cited Papers and 5 ESI Hot Papers, he has made significant contributions. Furthermore, his three papers are featured in the UK Research Excellence Framework (REF) 2021. Prof. Zhang has led many successful industrial projects and secured academic grants from NIH, Royal Society, GCRF, EPSRC, MRC, British Council, and NSFC. He has given over 120 invited talks at international conferences, universities, and companies and has served as (Co-)Chair for more than 60 international conferences and workshops.
Introduction of the Interviewer:
Dr. Caterina Giannitto, an excellent Junior Editorial Board member of JCMT, comes from Humanitas Research Hospital in Italy. Dr. Giannitto's research interests include head and neck oncology and imaging artificial intelligence.
Here are the details of the interview:
Q1. To start with, could you tell us a little about yourself and what you have done so far in deep learning?
My primary research interests revolve around the dynamic intersection of deep learning and medical image analysis. Throughout my academic and professional journey, I have been ardently engaged in pioneering advancements at the nexus of these two domains.
In deep learning, I have dedicated my efforts to developing novel theories and cutting-edge neural network architectures. These endeavors have broadened our understanding of the underlying principles governing deep learning and contributed significantly to the evolution of sophisticated models capable of handling complex medical image datasets.
Much of my research endeavors have been devoted to medical image classification's challenging yet pivotal task. By leveraging state-of-the-art deep learning techniques, I have strived to enhance the accuracy and efficiency of diagnostic processes. This involves training models to discern intricate patterns and subtle nuances within medical images, leading to more precise and timely identifications of various pathologies.
Furthermore, my expertise extends to the realm of medical image segmentation, where the objective is to delineate and characterize specific regions of interest within complex anatomical structures. I have delved into the development of advanced segmentation algorithms that leverage the power of deep learning to achieve remarkable precision in extracting relevant information from diverse medical imaging modalities.
Q2. Do you think that it is important to share your passion and knowledge?
I believe it is crucial to share both passion and knowledge. Passion serves as a powerful catalyst for inspiration, and by sharing it, we have the potential to ignite the enthusiasm of others, fostering a collaborative environment. Similarly, sharing knowledge is the cornerstone of progress. In a globally interconnected society, the free exchange of ideas accelerates innovation and collective learning.
By openly sharing our insights and expertise, we contribute to a culture of continuous growth, empowering individuals and communities to build upon shared knowledge for the betterment of society as a whole. In essence, sharing passion and knowledge enriches personal development and plays a vital role in shaping a more informed and collaborative future.
Q3. What is inside Deep learning?
Deep learning, a subfield of machine learning, encompasses a set of advanced algorithms designed to mimic the complex neural networks of the human brain. Deep learning involves using artificial neural networks with multiple layers (deep neural networks) to process and analyze data. These layers, comprising interconnected nodes or neurons, allow the system to learn hierarchical representations of features from raw input data. Deep learning architectures, such as convolutional neural networks (CNNs) for image analysis or recurrent neural networks (RNNs) for sequential data, excel at automatic feature extraction and pattern recognition.
The strength of deep learning lies in its ability to automatically discover intricate patterns and representations from vast datasets, enabling it to excel in tasks like image and speech recognition, natural language processing, and medical diagnosis.
Techniques like backpropagation, resilient backpropagation (Rprop), root mean square propagation (RMSprop), gradient descent, adaptive moment estimation (ADAM), etc., are commonly employed to train these deep networks, optimizing their parameters to enhance predictive accuracy. Deep learning transforms raw data into meaningful representations through layered neural networks, unleashing its potential across a broad spectrum of applications.
Q4. We want to highlight the importance of deep learning to the audience. How do you think deep learning will redefine medical image analysis and create a massive impact?
Deep learning is poised to revolutionize medical image analysis, ushering in a new era of precision and efficiency in healthcare. Its ability to automatically extract intricate patterns and features from complex medical images allows for unparalleled accuracy in diagnosis and prognosis. Deep learning algorithms, particularly convolutional neural networks, transfer learning, visual transformers (ViTs), etc., excel in tasks such as image classification and segmentation, enabling rapid and precise identification of anomalies within medical scans. These transformative technologies promise to significantly reduce the time required for diagnostics, enhance the detection of subtle abnormalities, and ultimately improve patient outcomes.
Integrating deep learning in medical image analysis streamlines the diagnostic process. It opens avenues for personalized medicine, where treatment plans can be tailored based on individualized insights derived from comprehensive image data. As deep learning continues to evolve, its impact on medical image analysis is poised to be massive, offering unprecedented advancements in the field and reshaping the landscape of healthcare by providing clinicians with powerful tools to make more informed decisions and ultimately save lives.
Q5. Artificial Neural networks have been studied for 50 years, but only recently have they achieved remarkable successes in such difficult tasks as speech and image recognition, with Deep Learning Networks. What factors enabled this success - big data, algorithms, hardware?
The recent remarkable successes of Artificial Neural Networks (ANNs), particularly with Deep Learning Networks (DLNs), can be attributed to various factors, each playing a crucial role in their newfound efficacy.
Firstly, the availability of vast amounts of labeled data, commonly called big data, has been instrumental. Deep learning models thrive on large datasets as they can automatically learn intricate patterns and representations, allowing for more robust generalization to diverse and complex tasks.
Secondly, algorithmic advancements have been pivotal, such as developing deeper architectures like convolutional neural networks (CNNs) and recurrent neural networks (RNNs). These sophisticated algorithms, coupled with improvements in training techniques like backpropagation and regularization, enable extracting hierarchical features from data, enhancing the models' ability to understand and represent complex relationships.
Finally, significant strides in hardware capabilities, including the advent of powerful Graphics Processing Units (GPUs) and specialized hardware for deep learning tasks, have accelerated the training of complex models. The synergy of big data, advanced algorithms, and enhanced hardware has propelled the success of deep learning networks, unlocking their potential in challenging tasks such as speech and image recognition, marking a transformative milestone in the field after decades of neural network research.
Q6. Since these technologies didn't really exist right from the beginning, do you think it is important for professionals who have already settled into their careers, to reskill/ upskill themselves in one of them?
The rapid evolution of technologies, especially in fields like artificial intelligence, machine learning, and deep learning, underscores the importance of continuous learning and adaptation for professionals who have already settled into their careers. The dynamic nature of these fields means that staying current with the latest advancements is beneficial and often necessary to remain competitive and relevant in the workforce.
Reskilling or upskilling in these emerging technologies equips professionals with the knowledge and tools to tackle evolving challenges, contribute to innovative solutions, and potentially open new career opportunities.
Moreover, as these technologies become increasingly integrated into various industries, professionals with a solid understanding of artificial intelligence, machine learning, and deep learning can offer valuable insights and contribute meaningfully to projects and initiatives that leverage these cutting-edge tools. In essence, fostering a mindset of continuous learning and adaptability is crucial for professionals to navigate the ever-changing landscape of technology and stay at the forefront of their respective fields.
Q7. Deep learning is not an easy method to use. What tools and tutorials would you recommend to learn more about it and use it on their data?
For those looking to delve into deep learning, leveraging the resources available to streamline the learning process is essential. To gain a solid foundation, I recommend starting with online courses and tutorials on platforms like Coursera, edX, and Khan Academy, where renowned experts often break down complex concepts into digestible modules.
Books such as "Deep Learning" by Ian Goodfellow and Yoshua Bengio offer comprehensive insights into the theoretical underpinnings. Platforms like TensorFlow and PyTorch for practical implementation provide powerful and widely used libraries for building and training deep learning models. Online communities, including forums like Stack Overflow and dedicated subreddits, are invaluable for troubleshooting issues and seeking guidance from the global deep-learning community.
Additionally, cloud-based platforms like Google Colab and Microsoft Azure offer accessible environments to experiment with deep learning on diverse datasets without the need for high-end hardware.
Ultimately, combining theoretical understanding from courses, hands-on experience with libraries like TensorFlow or PyTorch, and active participation in online communities can significantly ease the learning curve and empower individuals to use deep learning on their data effectively.
Q8. What are the challenges of deep learning in cancer imaging?
Deep learning in cancer imaging presents both immense potential and notable challenges. One significant challenge lies in the need for large and diverse datasets for training deep learning models effectively. Obtaining such datasets with comprehensive annotations for various cancer types can be time-consuming and resource-intensive. Another hurdle is the interpretability of deep learning models, as the complexity of neural networks makes it challenging to understand the rationale behind their predictions, which is crucial in medical decision-making. Another concern is ensuring the robustness and generalization of models across different populations and imaging modalities.
Additionally, addressing ethical considerations, patient privacy, and regulatory compliance is crucial in the healthcare domain. Finally, integrating deep learning models into clinical workflows requires meticulous validation and collaboration between computer scientists and medical professionals to ensure that these technologies provide meaningful and reliable support for cancer diagnosis, prognosis, and treatment planning. Despite these challenges, the potential impact of deep learning in cancer imaging underscores the importance of ongoing research and collaboration to overcome these obstacles and unlock the full benefits of this transformative technology in healthcare.
Q9. Can AI experience hallucinations? How do we identify false information generated by neural networks?
AI, particularly neural networks, does not experience hallucinations in the human sense, as it lacks subjective consciousness or perception. However, AI models can produce outputs that might be interpreted as hallucinatory or unrealistic.
Identifying false information generated by neural networks is a crucial concern in AI. One approach involves implementing robust validation techniques during training to ensure that models generalize well to various inputs and do not generate spurious outputs. Post-training, thorough testing with diverse datasets and real-world scenarios is essential to evaluate the model's reliability.
Additionally, anomaly detection methods and adversarial testing can be employed to identify instances where neural networks might generate misleading or inaccurate information. Transparency and interpretability tools, such as attention mechanisms, can provide insights into which parts of the input data the model focuses on during decision-making, aiding in identifying potential errors or hallucinatory outputs. Continued research into AI ethics and the development of trustworthy AI systems are crucial to mitigate the risks associated with false information generated by neural networks and ensure responsible deployment of these technologies across various domains.
Q10. What's next? What are the biggest hurdles to overcome and what are you most excited about in AI development?
Looking ahead in AI development, the field faces several significant challenges and promising opportunities. One prominent hurdle is achieving a deeper understanding and interpretability of complex neural networks, enhancing their explainability to build trust and facilitate their integration into critical domains like healthcare.
Additionally, addressing biases in AI algorithms regarding training data and model outputs is essential for creating fair and equitable systems. The ethical deployment of AI, focusing on privacy and security, remains a priority. On a positive note, the advancement of AI in solving real-world problems, such as drug discovery, climate modeling, and healthcare diagnostics, is fascinating.
The potential for AI to drive meaningful societal impact, coupled with ongoing research into mitigating its challenges, inspires optimism. As AI development progresses, interdisciplinary collaboration, ethical considerations, and a commitment to addressing societal concerns will be pivotal in realizing the full potential of artificial intelligence in shaping a better future.
The ongoing Special Issue "Deep Learning Applications in Cancer Research" led by Prof. Yudong Zhang is calling for high-quality papers! If you are interested in this international project, please submit your article to the following link:
Online submission
link: https://oaemesas.com/login?JournalId=jcmt&SpecialIssueId=jcmt20230705
Editor: Pan Wang
Language Editor: Catherine Yang
Production Editor: Yan Zhang
Respectfully Submitted by the Editorial Office of Journal of Cancer Metastasis and Treatment
