In Surgical Oncology, Radiomics: Challenges and Applications

Ellen Bock^*
*Correspondence: Ellen Bock, Department of Surgery, University Medical Center Utrecht, Utrecht, The Netherlands, Email:

Keywords

Tumors • Radiomics • Neoadjuvant

Introduction

We used both the tumor and ALN radiomic features to create the ALN tumor radiomic signature in light of the link between ALN and the primary tumor site. This resulted in a higher AUC than when we used ALN or tumor radiomic features alone. More importantly, the predictive power may be enhanced by combining ALN and tumor radiomic features with clinical, pathologic, and molecular subtype characteristics. In our model, independent validation cohorts were tested to ensure the competence of the constructed signatures. In addition, our findings revealed associations between MRI radiomic features and tumor microenvironment characteristics like immune cells, long non- coding RNAs, and types of methylated sites, as well as significant differences in MRI radiomics before and after neo-adjuvant chemotherapy [1].

Literature Review

Computer software extracts a large number of predefined high throughput features from radiomics to characterize lung lesions using Artificial Intelligence (AI). Statistical methods then filter the features that are most relevant to the results. Finally, a diagnostic and predictive model is developed using machine learning. At the moment, a lot of attention is being paid to how radiomics can be used to diagnose and evaluate pulmonary nodules. The significance of texture in the diagnosis of pGGNs was examined by Hwang, et al. However, due to the small number of cases (64), fewer features were extracted. A radiomics based model performed better in our previous studies for predicting sub-centimeter GGNs; however, this study also included pGGNs and part solid nodules [2]. The diagnosis of part solid nodules is straight forward due to the satisfactory correlation of the solid components on CT imaging with pathologically invasive components; however, the diagnosis of pGGNs is challenging and the invasiveness is frequently underestimated. In order to increase the diagnostic accuracy of pGGNs and provide a foundation for rational clinical decision making, the goal of this study was to develop a comprehensive model of pGGNs based on conventional CT features and radiomics.

The effectiveness of a treatment can be evaluated using criteria from pathology or radiology. A "hard" endpoint is the pathologic response; however, only 16% of NSCLC patients who undergo surgical resection can have it evaluated. Therefore, the primary component of evaluating NSCLC treatment response is radiologic response. Based on a one-dimensional evaluation of the tumor's size, the Response Evaluation Criteria in Solid Tumors (RECIST) provides a standardized, objective method for reporting therapy response [3].

Discussion

Oncology trial endpoints like response rate and progression-free survival are defined using RECIST criteria. The radiologic evaluation of treatment response in clinical practice largely relies on tumor size, with a qualitative assessment of other tumor characteristics like homogeneity and shape added in. In order to solve this problem, we created Pyradiomics, a comprehensive open-source platform that makes it possible to process and extract radiomic features from medical image data by employing a wide range of engineered, hard-coded feature algorithms. In addition to a back-end interface that enables automation of data processing, feature definition, and batch handling, pyradiomics provides a flexible analysis platform with a front end interface in 3D slicer, a free open-source platform for medical image computing. Pyradiomics can be installed on any system because it is implemented in python, a well-known open-source language for scientific computing. [4].

In epithelial cell lines, EGFR is a receptor tyrosine kinase that controls normal cell growth. About 40% of glioblastomas have EGFR mutations, but lower grade gliomas rarely have them. The splice variant III (EGFRvIII) is the most common extracellular EGFR mutation in glioblastoma, occurring in 31% of patients. There is a moderate relationship between EGFR amplification and tumor blood flow and volume in radiogenomic studies using perfusion imaging. Approaches based on support vector machines have demonstrated that EGFRvIII mutant gliomas have deep peritumoral infiltration, indicating a more aggressive or infiltrative phenotype. Neovascularization, cell density, and a preference for the frontal and parietal regions in EGFRvIII tumors have all been demonstrated by additional multivariate approaches employing a larger number of multi-parametric features [5].

Data mining, which is the process of finding patterns in large data sets, can be done with large, high-quality, and carefully curated data sets that are available. Artificial intelligence, machine learning, or statistical methods can be used in this process. On one end, there are supervised and unsupervised methods like neural networks, support vector machines, and Bayesian networks for machine learning. Even though these methods make use of a priori knowledge by training sets, they are agnostic in the sense that they do not make any assumptions about what each feature means. As a result, all features are given equal weight when learning starts. Hypothesis driven methods that group features according to predetermined information content are at the other end of the data mining spectrum [6]. The best models are those that are tailored to a specific medical context and, as a result, start with a well-defined endpoint. Both of these approaches have merit.

Conclusion

Radiomics have shown extraordinary commitment in customizing chemotherapy treatment and choices for patients. The studies examined in this paper have demonstrated the usefulness of radiomics and machine learning in determining which patients benefit from chemotherapy and which do not. To make radiomic analysis more widely used and effective in clinical settings, stronger auto-segmentation algorithms, increased data sharing, and standardization are required.

Acknowledgement

None.

Conflict of Interest

None.

References

1. Masuda, Norikazu, Soo-Jung Lee, Shoichiro Ohtani and Young-Hyuck Im, et al. "Adjuvant capecitabine for breast cancer after preoperative chemotherapy." New Eng J Med 376 (2017): 2147-2159.

Google Scholar, Crossref, Indexed at

2. Pfob, André and Peter Dubsky. "The underused potential of breast conserving therapy after neoadjuvant system treatment–Causes and solutions." Breast 67 (2023): 110-115.

Google Scholar, Crossref, Indexed at

3. Minella, Carola, Andrea Villasco, Marta D’Alonzo and Lisa Cellini, et al. "Surgery after neoadjuvant chemotherapy: a clip-based technique to improve surgical outcomes, a single-center experience." Cancers 14 (2022): 2229.

Google Scholar, Crossref, Indexed at

4. Pastorello, Ricardo G., Alison Laws, Samantha Grossmith and Claire King, et al. "Clinico-pathologic predictors of patterns of residual disease following neoadjuvant chemotherapy for breast cancer." Modern Pathol 34 (2021): 875-882.

Google Scholar, Crossref, Indexed at

5. Thomassin-Naggara, Isabelle, Lucie Lalonde, Julie David, Emile Darai, Serge Uzan, and Isabelle Trop. "A plea for the biopsy marker: how, why and why not clipping after breast biopsy?" Breast Cancer Res Treat 132 (2012): 881-893.

Google Scholar, Crossref, Indexed at

6. Sever, Ali R., Mary ER O’Brien, Stephen Humphreys and Inderjit Singh, et al. "Radiopaque coil insertion into breast cancers prior to neoadjuvant chemotherapy." Breast 14 (2005): 108-117.

Google Scholar, Crossref, Indexed at