Written By Victoria Salazar on April 10, 2023

Introduction

The use of Artificial Intelligence (AI) in the field of medical imaging has seen remarkable progress in recent years. AI-powered algorithms are capable of analyzing large amounts of data with speed and accuracy, making them valuable tools for diagnosing and predicting diseases, including cancer. In diagnostic radiology, AI is used to enhance cancer detection and improve patient outcomes. However, there are still challenges to overcome in the use of AI for cancer detection in diagnostic radiology. 

Advancements to AI for Cancer Detection in Diagnostic Radiology

Computer-Aided Diagnosis (CAD) Systems

Computer-Aided Diagnosis (CAD) systems are AI-based algorithms that aid radiologists in analyzing medical images for the detection of cancer. CAD systems work by analyzing the texture, shape, and intensity of medical images to identify regions that may be indicative of cancer. These regions are then flagged for further analysis by radiologists, who can make a more informed diagnosis based on the system's findings. CAD systems have been shown to be an effective tool for improving the accuracy of cancer detection in diagnostic radiology, particularly in mammography and lung cancer screening.

Mammography is a widely used screening method for breast cancer detection. However, mammography has a high false-positive rate, leading to unnecessary biopsies and increased patient anxiety. CAD systems have been shown to improve mammography's sensitivity and specificity by reducing false-positive rates and increasing the accuracy of cancer detection. A study by Lehman et al. (2015) found that the use of CAD systems increased the sensitivity of mammography by 4.2% and reduced false-positive rates by 15.9%. Similarly, CAD systems have been shown to improve lung cancer screening by detecting pulmonary nodules that may be indicative of cancer. A study by Brown et al. (2007) found that CAD systems increased the sensitivity of lung cancer detection by 20%, leading to earlier detection and improved patient outcomes.

Radiomics

Radiomics is an emerging field that involves the analysis of medical images using AI algorithms to extract quantitative data on tumor phenotype, microenvironment, and response to treatment. Radiomics aims to provide a non-invasive way of predicting treatment response and patient outcomes in cancer patients, which can be used to develop personalized treatment plans for patients. Radiomics involves the extraction of features from medical images, such as shape, texture, and intensity, which are then analyzed using machine learning algorithms to identify patterns and correlations between imaging features and clinical outcomes. Radiomics has shown to be a promising tool for improving cancer detection and predicting patient outcomes. A study by van Timmeren et al. (2019) found that radiomics analysis of CT scans improved the accuracy of lung cancer diagnosis and prediction of patient outcomes. The study showed that radiomics features extracted from CT scans could predict treatment response and overall survival in lung cancer patients.

The use of radiomics in cancer detection and treatment planning is still in its early stages, but it has already shown promising results in various cancer types, including lung, breast, and prostate cancer. For instance, a study by Antropova et al. (2019) demonstrated that radiomics analysis of mammograms could improve the accuracy of breast cancer diagnosis and reduce false positives. Another study by Langer et al. (2019) found that radiomics analysis of MRI scans could predict biochemical recurrence in prostate cancer patients after surgery.

Despite the potential benefits of radiomics in cancer detection and treatment, there are also some challenges associated with this approach. One of the key challenges is the lack of standardization in radiomics analysis, which can lead to inconsistencies in the results obtained from different studies. Additionally, the interpretation of radiomics features can be complex, and there is a need for more research to validate the clinical relevance of these features. Radiomics has the potential to revolutionize cancer diagnosis and treatment by providing a more comprehensive understanding of the tumor and its response to treatment. As AI algorithms continue to improve, radiomics is likely to become an increasingly important tool in the fight against cancer.

Challenges to AI for Cancer Detection in Diagnostic Radiology

Interpretability

Radiologists and clinicians need to be able to interpret the results of AI algorithms to make informed decisions about patient care. The lack of interpretability of AI algorithms is a significant barrier to their adoption in clinical practice. AI algorithms are often viewed as black boxes, making it difficult to understand how they arrive at their decisions or the reasoning behind an AI algorithm’s output. If clinicians cannot understand how an AI algorithm arrived at a particular diagnosis or decision, they may be hesitant to use it in practice. This lack of trust can ultimately limit the adoption and implementation of AI systems for cancer detection. The lack of AI interpretability, for this reason, has high stakes and consequences, especially if a missed diagnosis can be severe. Developing more interpretable AI algorithms for cancer detection is crucial for their adoption in clinical practice and for improving patient outcomes. 

However, interpretability in AI can be improved through the use of explainable AI (XAI) techniques. XAI techniques aim to make AI algorithms more transparent by providing interpretable outputs and explanations of how the algorithms arrived at their decisions. XAI techniques, such as feature importance analysis and decision tree visualization, can help radiologists and clinicians to understand the factors that contribute to AI algorithms' decisions.

Data Quality

The accuracy and quality of the data used to train AI algorithms directly impact the algorithms' performance and effectiveness. Inaccurate or incomplete data can lead to biased and erroneous AI algorithms that may result in incorrect diagnoses and patient harm.

Data quality can be improved through the use of standardized imaging protocols and the collection of diverse and representative data. Standardized imaging protocols ensure that medical images are acquired consistently across different imaging modalities and institutions, reducing variability and increasing data quality. The collection of diverse and representative data ensures that AI algorithms are trained on a wide range of imaging data, reducing the risk of bias and improving the algorithms' generalizability to new patient populations.

Conclusion

AI has shown tremendous promise in enhancing cancer detection and improving patient outcomes in diagnostic radiology. Advancements such as CAD systems and Radiomics have demonstrated great potential in improving cancer detection accuracy and predicting patient outcomes. However, challenges such as interpretability and data quality must be addressed to improve the adoption and effectiveness of AI in clinical practice. Explainable AI techniques and standardized imaging protocols can improve interpretability and data quality, respectively. As AI continues to evolve, it will be essential to address these challenges to fully realize the potential of AI in cancer detection and patient care.

References:

1. Brown, M. S., Goldin, J. G., Rogers, S. R., & McNitt-Gray, M. F. (2007). Automated identification of pulmonary nodules in helical CT images using morphological techniques. In Medical Imaging 2007: Image Processing (Vol. 6512, p. 65123D). International Society for Optics and Photonics.

2. Lehman, C. D., Wellman, R. D., Buist, D. S., Kerlikowske, K., Tosteson, A. N., Miglioretti, D. L., ... & Stout, N. K. (2015). Diagnostic accuracy of digital screening mammography with and without computer-aided detection. JAMA internal medicine, 175(11), 1828-1837.

3. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278(2):563-577.

4. van Timmeren, J. E., Leijenaar, R. T., van Elmpt, W. J., Reymen, B., Monshouwer, R., Bussink, J., ... & Lambin, P. (2019). Survival prediction of non-small cell lung cancer patients using radiomics analyses of cone-beam CT images. Radiotherapy and Oncology, 129(1), 17-24.

5. van Timmeren JE, Cester D, Tanadini-Lang S, et al. Radiomics in medical imaging-“how-to” guide and critical reflection. Insights into Imaging. 2020;11(1):91. doi:10.1186/s13244-020-00868-0.

6. Zhang, J., & Wang, S. (2018). Challenges and opportunities of artificial intelligence in radiology. Medical physics, 45(6), e961-e981.