Nasopharyngeal Carcinoma Post-therapy PET/CT Images

Nuclear medicine
PET/CT images
Nuclear Medicine

The researches find that the radiomics-based differentiation of local recurrence versus inflammation from post-treatment nasopharyngeal positron emission tomography/X-ray computed tomography PET/CT images The standard-of-care treatment for nasopharyngeal carcinoma (NPC) is radiotherapy with or without chemotherapy. Despite significant improvement in local control due to improved treatment techniques, local recurrence still accounts for about 28 % of NPC cases; and nearly 13 % of
patients die of local relapse.

Tomography PET/CT images

As such, early diagnosis and accurate identification of local recurrence is essential for timely implementation of salvage treatment. However, the diagnosis of local recurrence is commonly affect by post-therapy changes of nasopharyngeal tissues, such as edema, inflammation, fibrosis, and scarring. Anatomic imaging modalities; such as CT and MRI scans, have been report to be ineffective in differentiating recurrent/ residual tumors from inflammatory tissue after treatment.

2-Deoxy-2-[18F]fluoro-D-glucose ([18F]FDG) positron emission tomography/X-ray computed tomography (PET/CT) imaging has been established as a powerful technique in the management of NPC. However, [18F]FDG PET/CT exhibits accumulated uptake in both malignant and benign masses, and is not tumor-specific. In particular, increased glucose metabolism of inflammatory tissue makes the differentiation of recurrence more challenging.

A powerful technique

Conventional metrics (such as SUV max /mean /peak, metabolically active tumor volume (MATV), and total lesion glycolysis (TLG)) have been widely used in PET/CT clinical oncology. Such metrics cannot describe the complex spatial distribution of malignant tumors, which is often referr as intra-tumor heterogeneity. The quantification of intra-tumor heterogeneity may provide additional information about the tumor phenotype. The emerging field of radiomics provides comprehensive
quantification of intra-tumor heterogeneity by extracting and mining a large number of radiomics features.

The potential of radiomics to improve clinical decision-making (i.e., diagnosis, prognosis, treatment response assessment)
has been extensively investigated, and radiomics modelling has demonstrated strong prediction performance across a range of cancer types. Radiomics modelling involves three major steps:

Gray-level discretization

Feature selection, model building/ testing, and validation. Considering the numerous parameters impacting radiomics features (e.g., reconstruction settings, segmentation methods, gray-level discretization methods and bin sizes), the possible number of radiomics features is virtually unlimited and causes overfitting. Feature selection is a crucial step to identify useful, robust, and non-redundant features and to avoid overfitting.

For model building, multiple machine learning methods have demonstrated different prediction performances for same context validations. The identification of optimal machine learning methods is also another crucial step towards stable and high quality clinical decision-making. Therefore, the principle challenge of radiomics modelling is the optimal collection and integration of feature selection and modelling methodology.