Talk:Radiomics
Add topic | This article is rated Start-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | ||||||||||
| |||||||||||
Wiki Education assignment: ECE584 - Medical Imaging Systems
[edit]
This article was the subject of a Wiki Education Foundation-supported course assignment, between 9 March 2026 and 28 May 2026. Further details are available on the course page. Student editor(s): Rogersec, Youngye0707 (article contribs).
— Assignment last updated by Xuqi555801 (talk) 21:39, 3 May 2026 (UTC)
Proposed content for future consideration
[edit]@MathWizard88:A 2026 multi-author study from Mayo Clinic published in Gut—a BMJ journal—reported on a radiomics-based AI model (REDMOD) for early detection of pancreatic cancer on routine CT scans. The study described detection of pre-clinical disease signatures up to three years before diagnosis, with reported sensitivity of 73% compared to 39% for radiologist review.
Note: Per WP:MEDRS and WP:RECENTISM, this represents a single validation study and should not be treated as definitive or consensus medical knowledge. Independent replication, prospective clinical trials, and systematic review or meta-analysis would be needed before inclusion in the main article as established fact. These references are archived here for potential future use once the technology's clinical utility is more thoroughly established in the peer-reviewed literature.
I maintained the full list of researchers which will be too lengthy for the main page.Oceanflynn (talk) 15:13, 30 April 2026 (UTC)
References
- Mukherjee, Sovanlal; Antony, Ajith; Patnam, Nandakumar G; Trivedi, Kamaxi H.; Karbhari, Aashna; Bhinder, Khurram Khaliq; Zarrintan, Armin; Fletcher, Joel G.; Truty, Mark; Johnson, Matthew P; Chari, Suresh T.; Goenka, Ajit Harishkumar. "Next-generation AI for visually occult pancreatic cancer detection in a low-prevalence setting with longitudinal stability and multi-institutional generalisability" (PDF). Gut. The BMJ: 13. doi:10.1136/gutjnl-2025-337266. Retrieved 30 April 2026.
- Murphy, Susan (29 April 2026). "Mayo Clinic AI detects pancreatic cancer up to 3 years before diagnosis in landmark validation study". Mayo Clinic News Network. Retrieved 30 April 2026.
- Gale, Jason (29 April 2026). "AI Spots Pancreatic Cancer Years Before It Shows Up, Study Finds". Bloomberg News. Retrieved 30 April 2026.
Proposed Limitations Section
[edit]I noticed that although the article lists some of the challenges in radiomics, it doesn't have a clear section summarizing the limitations of the field or the efforts to address them. I thought it might be beneficial to add a short "Limitations" subsection (and possibly a brief "Future directions subsection) near the applications section. A current draft looks like:
Limitations
Radiomics faces a few challenges that limits its ability to be widely implemented in real clinical applications [1]. The first issue occurs prior to the input of medical images, existing in the collection of training data. There is no standardization in image protocols, so medical facilities use a variety of image parameters, contrast application, equipment, and techniques [1]. Radiologists also manually segment images which imposes subjectivity into the process that can largely impact the extracted features [1][2]. Imaging data from prior scans can be easily obtained, but without careful selection of the input images, AI models will be susceptible to unreliable training data that affects the performance of the predictive model [3].
Due to there being more extracted features in the data than number of samples, the dataset will be high-dimensional [2]. The challenges with the resulting highly correlated training set lies in the danger of overfitting [2]. The model will learn to identify the noise of the data instead of reliable biological markers, resulting in a model that looks to have good performance until it sharply declines when introduced to independent validation [2][3]. Even if the model was able to perform well outside of the training data, it would be challenging for clinicians to trust due to it being difficult to understand the how the resulting model works [2].
Future Directions
Efforts to improve standardization of imaging protocols and imaging preprocessing are being made [1]. This, along with obtaining a diverse set of images, would drastically improve the performance of the models [2]. As Radiomics becomes more popular, larger sets of data will be available which could help the high correlation issues in current models [2]. Using multi-institutional data will also eliminate training bias to ensure that models will perform on new patient populations [3].
References
[edit]- 1 2 3 4 Qi, Ying-Jia, et al. "Radiomics in Breast Cancer: Current Advances and Future Directions." Cell Reports Medicine, vol. 5, no. 9, 2024, doi:10.1016/j.xcrm.2024.101719.
- 1 2 3 4 5 6 7 Demircioğlu, Aydın. "Reproducibility and Interpretability in Radiomics: A Critical Assessment." Diagnostic and Interventional Radiology, vol. 31, no. 4, 2025, pp. 321–328, doi:10.4274/dir.2024.242719.
- 1 2 3 Bera, Kaustav, et al. "Predicting Cancer Outcomes with Radiomics and Artificial Intelligence in Radiology." Nature Reviews Clinical Oncology, vol. 19, no. 2, 2022, pp. 132–146, doi:10.1038/s41571-021-00560-7.