Correlation of texture feature analysis with bone marrow infiltration in initial staging of patients with lymphoma using $^{18}F-fluorodeoxyglucose$ positron emission tomography combined with computed tomography

Purpose: To explore whether radiomic features of fluorine-18-fluorodeoxyglucose ($^{18}F-FDG$) positron emission tomography-computed tomography (PET/CT) has association with bone marrow infiltration (BMI) in comparison to other conventional PET metrics. Material and methods: Forty-four patients (with pathologically proven lymphoma disease) underwent staging $^{18}F-FDG PET/CT$ scan. Primary tumour was semi-automatically or manually segmented with a threshold standardised uptake value (SUV) of 3. A total of 73 features were extracted from eight different textures. Spearman correlation was used to test the correlation of features with conventional quantitative metrics such as SUV, metabolic tumour volume, and total lesion glycolysis. Specificity and sensitivity (including 95% confidence intervals [CI]) for each of the studied parameters were derived using receiver operative characteristic (ROC) curves. Univariate and multivariate analyses were used to identify independent predictors associated with BMI. Results: Correlation between conventional PET metrics and features ranged between 0.50 and 0.97 for positive correlation (33 significant association features) and ranged from -0.52 to -0.97 for inverse correlation (three significant association features) for both strong and moderate correlations. Analysis of ROC curves showed that high-intensity long-run emphasis 4 bin, high-intensity large zone emphasis 64 bin, long-run emphasis (LRE) 64 bin, large-zone emphasis 64 bin, max spectrum 8 bin, busyness 64 bin, and code similarity 32 and 64 bin were significant discriminators of BMI among other features (area under curve > 0.682, p < 0.05). Univariate analyses of texture features showed that code similarity and long-run emphasis (both 64 bin) were significant predictors of bone marrow involvement. Multivariate analyses revealed that LRE (64 bin, p = 0.031) with an odds ratio of 1.022 and 95% CI of (1.002-1.043) were independent variables for bone marrow involvement. Conclusions: $^{18}F-FDG PET/CT$ radiomic features are synergistic to visual assessment of BMI in patients diagnosed with lymphoma using $^{18}F-FDG PET/CT$. Further assessment of long-run emphasis is highly warranted