• info@icbcongress.com
• English

ساخت اکانت کاربری رایگان

• ورود

INTERNATIONAL CONGRESS OF

CancerGenetics

• Application of Artificial Intelligence in Cancer Radiology and Prediction of Tumor Surgical Outcomes: A Systematic Review of Radiomic Models and Multimodal Approaches

• Arshia Neshat Taherzadeh,^1,* Seyed Pouria Mirrezai,² Romina Tohidi,³
  1. Faculty of Veterinary Medicine, Islamic Azad University, Science and Research Branch, Tehran, Iran
  2. Faculty of Veterinary Medicine, Islamic Azad University, Garmsar Branch, Semnan, Iran
  3. Faculty of Veterinary Medicine, Islamic Azad University, Garmsar Branch, Semnan, Iran
  
  
  
• Introduction: Early and accurate cancer diagnosis along with reliable prediction of surgical outcomes is one of the most important challenges in modern oncology. Despite significant advances in imaging modalities such as CT, MRI, and PET, traditional interpretation of these images by radiologists faces limitations such as inter-observer variability, inability to detect microscopic sub-image patterns, and inability to predict tumor biological invasiveness. Artificial intelligence (AI), especially the branches of radiomics and deep learning, with the ability to extract hundreds of quantitative features from images beyond human comprehension, has created the opportunity to transform diagnostic imaging from a qualitative tool into a quantitative predictive platform. The aim of this systematic review is to comprehensively evaluate the utility of radiological image-based AI models in predicting surgical outcomes of tumors, including local recurrence, disease-free survival, microvascular invasion, surgical margin involvement, and response to neoadjuvant therapies.
• Methods: This systematic review was conducted according to the PRISMA guideline, and a comprehensive search of PubMed/MEDLINE, Scopus, Web of Science, Cochrane Library, and IEEE Xplore databases was performed from January 2020 until April 2026. The main keywords included a combination of "Artificial Intelligence", "Radiomics", "Deep Learning", "Cancer Imaging", "Surgical Outcome", "Recurrence Prediction", "Overall Survival", and "Margins Status" using Boolean operators. Inclusion criteria included original studies with prospective or retrospective designs, clinical trials, meta-analyses, and systematic reviews that evaluated AI models based on radiological images (CT, MRI, PET/CT) to predict surgical outcomes in solid tumors (lung, liver, kidney, colorectal, breast, prostate, and brain). Studies without external validation, sample sizes of less than 50 patients, and animal studies were excluded. In total, after screening the titles/abstracts of 1,542 articles, 16 studies met the inclusion criteria and were included in the final synthesis.
• Results: Research has shown that AI can predict whether cancer will return after surgery by looking at CT and MRI images of patients before surgery. For example, AI models have been able to predict cancer recurrence in patients with kidney and lung tumors with about 87 to 90 percent accuracy. The technology can also detect whether cancer cells have invaded blood vessels around the tumor (which increases the risk of recurrence) with about 87 percent accuracy. In liver cancer, AI can even detect the aggressiveness of the tumor (a cell proliferation marker called Ki-67) from the images with 92 percent accuracy. Importantly, when AI combines radiology images with pathology information from tumor tissue, it provides more accurate predictions than when using only one data source. However, most of these studies have been conducted in a laboratory setting and still require further validation for routine use in hospitals.
• Conclusion: This systematic review demonstrates that radiological cancer-based AI has made significant progress in predicting surgical outcomes of tumors over the past five years. By extracting sub-image features that are not discernible by the human eye, radiomics and deep learning models have been able to predict local recurrence, microvascular invasion, and cell proliferation markers with high accuracy. The most important clinical achievement is the potential of these models to identify high-risk patients before surgery, which allows for the design of personalized treatment strategies.
• Keywords: Radiomics, Surgical outcome prediction, Cancer imaging, Multimodal AI fusion, Solid tumors