Artificial intelligence applications on restaging [18F]FDG PET/CT images of metastatic colorectal cancer

Introduction: to investigate the application of [¹⁸F]FDG PET/CT images-based textural features analysis to propose a radiomics models able to early predict disease progression and survival outcome in metastatic colon cancer patients after first-line treatments.

Methods: We retrospectively analyzed fifty-two colorectal metastatic patients who underwent restaging [¹⁸F]FDG PET/CT during the restaging process of the disease after first-line therapy. Follow-up data were recorded for a minimum of 12 months after the PET/CT scan. PET/TC images were imported in LIFEx toolbox to extract 105 and 66 features from each avid lesion in PET and low-dose CT images, respectively. A statistical system based on correlation matrix and point-biserial-correlation coefficient has been implemented for features reduction and selection, while Discriminant analysis (DA) was used as a predictive model. In order to assess the performance of the features in predicting progression disease we performed: per lesion analysis (only PET dataset; PET/CT dataset) ; per patient analysis (only PET dataset, PET/CT dataset); only liver lesions analysis (only PET dataset, PET/CT dataset). All lesions were again considered to assess the diagnostic performance of the features in discriminating only liver lesions.

Results: In predicting progression disease in whole group of patients, on PET features radiomics analysis, among per lesion analysis, the feature selected as more accurate for the DA classifier was GLZLM_GLNU, while 3 features (GLZLM_ZLNU and GLRLM_SRHGE among the CT features and GLZLM_GLNU among the PET features) were selected from PET/CT images dataset obtaining a mild empowerment of the accuracy when CT features analysis was merged with PET features performances (AUROC 65.22%). In per patient analysis for only-PET images, 3 features (GLZLM_ZLNU, GLZLM_HGZ, CONVENTIONAL_RIM_SUVbwstdev2) and 1 feature (CONVENTIONAL_HUKurtosis) considering the PET/CT dataset were selected by DA classifier (AUROC 61%). Performances in prediction of disease outcome by defining at liver per-lesion analysis one PET feature (GLZLM_GLNU AUROC 39.94%) from only-PET images and three PET/CT features (GLZLM_ZLNU, and GLRLM_SRHGE between the CT features and GLZLM_GLNU between the PET features with AUROC 55.26%) were identified. Similarly, in liver lesions per patient analysis, we found three PET features (GLZLM_ZLNU, GLZLM_HGZ, CONVENTIONAL_RIM_SUVbwstdev2 with AUROC 60.11%), and a PET/CT feature (CONVENTIONAL_HUKurtosis with AUROC 43.48%). In discrimination of liver metastasis from the rest of the other lesions optimal results of only-PET imaging were found for one feature (DISCRETIZED_SUVbwmin; AUROC 88.91%) and two features for merged PET/CT features analysis (DISCRETIZED_HISTO_Energy between the CT features and DISCRETIZED_SUVbwmin between the PET features; AUROC 95.33%).

Conclusions: Our machine learning model on restaging [¹⁸F]FDG PET/CT demonstrated to be feasible and useful in the predictive evaluation of disease progression in metastatic colon cancer after first line therapies. New investigations might propose morpho-functional based radiomics algorithms for risk stratification and impact on treatment management in colorectal cancer.