Effect of machine learning re-sampling techniques for imbalanced datasets in 18F-FDG PET-based radiomics model on prognostication performance in cohorts of head and neck cancer patients.

PURPOSE: Biomedical data frequently contain imbalance characteristics which make achieving good predictive performance with data-driven machine learning approaches a challenging task. In this study, we investigated the impact of re-sampling techniques for imbalanced datasets in PET radiomics-based prognostication model in head and neck (HNC) cancer patients. METHODS: Radiomics analysis was performed in two cohorts of patients, including 166 patients newly diagnosed with nasopharyngeal carcinoma (NPC) in our centre and 182 HNC patients from open database. Conventional PET parameters and robust radiomics features were extracted for correlation analysis of the overall survival (OS) and disease progression-free survival (DFS). We investigated a cross-combination of 10 re-sampling methods (oversampling, undersampling, and hybrid sampling) with 4 machine learning classifiers for survival prediction. Diagnostic performance was assessed in hold-out test sets. Statistical differences were analysed using Monte Carlo cross-validations by post hoc Nemenyi analysis. RESULTS: Oversampling techniques like ADASYN and SMOTE could improve prediction performance in terms of G-mean and F-measures in minority class, without significant loss of F-measures in majority class. We identified optimal PET radiomics-based prediction model of OS (AUC of 0.82, G-mean of 0.77) for our NPC cohort. Similar findings that oversampling techniques improved the prediction performance were seen when this was tested on an external dataset indicating generalisability. CONCLUSION: Our study showed a significant positive impact on the prediction performance in imbalanced datasets by applying re-sampling techniques. We have created an open-source solution for automated calculations and comparisons of multiple re-sampling techniques and machine learning classifiers for easy replication in future studies.

Evaluation of Thalassaemia Screening Tests in the Antenatal and Non-Antenatal Populations in Singapore.

INTRODUCTION: Haemoglobinopathy testing is performed for carrier screening and evaluation of microcytic anaemia. We evaluated the effectiveness of thalassaemia screening tests at our institution and suggest ways of improving the testing algorithm. MATERIALS AND METHODS: A total of 10,084 non-antenatal and 11,364 antenatal samples with alkaline gel electrophoresis (AGE), capillary electrophoresis (CE), haemoglobin H (HbH) inclusion test, mean corpuscular haemoglobin (MCH) and mean corpuscular volume (MCV) were retrospectively reviewed. A subgroup of 187 samples with genetic testing was correlated with HbH inclusions and MCH/ MCV. The effect of iron deficiency on percentage hemoglobin A2 (HbA2) was studied. RESULTS: HbH inclusion test showed low sensitivity of 21.43% for α-thalassaemia mutations but higher sensitivity of 78.95% for --SEA deletion. By receiver operating characteristic (ROC) analysis, MCH ≤28 pg or MCV ≤80 fl for non-antenatal samples and MCH ≤27 pg or MCV ≤81 fl for antenatal samples had >98% sensitivity for HbH inclusions. Above these thresholds, the probability that HbH inclusions would be absent was <99% (negative predictive value [NPV] >99%). MCH ≥28 pg had 100% sensitivity (95% CI 95.63%-100%) for α-thalassaemia mutations and 97.68% calculated NPV in the antenatal population. Detection of haemoglobin variants by CE correlated highly with AGE (99.89% sensitivity, 100% specificity). Severe iron deficiency reduced HbA2 in hemoglobin (P <0.001) and α-thalassaemia (P = 0.0035), but not in ß-thalassaemia. CONCLUSION: MCH/MCV thresholds have adequate sensitivity for α-thalassaemia in the antenatal population, and genotyping plays an important role as HbH inclusion test shows low sensitivity. CE without AGE, may be used as initial screening for haemoglobin variants. Our study provides contemporary data to guide thalassaemia screening algorithms in Singapore.

MRI texture features may predict differentiation and nodal stage of cervical cancer: a pilot study.

BACKGROUND: Texture analysis in oncological magnetic resonance imaging (MRI) may yield surrogate markers for tumor differentiation and staging, both of which are important factors in the treatment planning for cervical cancer. PURPOSE: To identify texture features which may predict tumor differentiation and nodal status in diffusion-weighted imaging (DWI) of cervical carcinoma. MATERIAL AND METHODS: Twenty-three patients were enrolled in this prospective, institutional review board (IRB)-approved study. Pelvic MRI was performed at 3-T including a DWI echo-planar sequence with b-values 40, 300, and 800 s/mm2. Apparent diffusion coefficient (ADC) maps were used for region of interest (ROI)-based texture analysis (32 texture features) of tumor, muscle, and fat based on histogram and gray-level matrices (GLM). All features confounded by the ROI size (linear model) were excluded. The remaining features were examined for correlations with histological differentiation (Spearman) and nodal status (Kruskal-Wallis). Hierarchical cluster analysis was used to identify correlations between features. A P value < 0.05 was considered statistically significant. RESULTS: Mean age was 55 years (range = 37-78 years). Biopsy revealed two well-differentiated, eight moderately differentiated, two moderately to poorly differentiated tumors, and five poorly differentiated tumors. Six tumors could not be graded. Lymph nodes were involved in 11 patients. Three GLM features correlated with the differentiation: LRHGE (ϱ = 0.53, P = 0.03), ZP (ϱ = -0.49, P < 0.05), and SZE (ϱ = -0.51, P = 0.04). Two histogram features, skewness (0.65 vs. 1.08, P = 0.04) and kurtosis (0.53 vs. 1.67, P = 0.02), were higher in patients with positive nodal status. Cluster analysis revealed several co-correlations. CONCLUSION: We identified potentially predictive GLM features for histological tumor differentiation and histogram features for nodal cancer stage.