Reproducibility in Radiomics: A Comparison of Feature Extraction Methods and Two Independent Datasets.

Radiomics involves the extraction of information from medical images that are not visible to the human eye. There is evidence that these features can be used for treatment stratification and outcome prediction. However, there is much discussion about the reproducibility of results between different studies. This paper studies the reproducibility of CT texture features used in radiomics, comparing two feature extraction implementations, namely the MATLAB toolkit and Pyradiomics, when applied to independent datasets of CT scans of patients: (i) the open access RIDER dataset containing a set of repeat CT scans taken 15 min apart for 31 patients (RIDER Scan 1 and Scan 2, respectively) treated for lung cancer; and (ii) the open access HN1 dataset containing 137 patients treated for head and neck cancer. Gross tumor volume (GTV), manually outlined by an experienced observer available on both datasets, was used. The 43 common radiomics features available in MATLAB and Pyradiomics were calculated using two intensity-level quantization methods with and without an intensity threshold. Cases were ranked for each feature for all combinations of quantization parameters, and the Spearman's rank coefficient, rs, calculated. Reproducibility was defined when a highly correlated feature in the RIDER dataset also correlated highly in the HN1 dataset, and vice versa. A total of 29 out of the 43 reported stable features were found to be highly reproducible between MATLAB and Pyradiomics implementations, having a consistently high correlation in rank ordering for RIDER Scan 1 and RIDER Scan 2 (rs > 0.8). 18/43 reported features were common in the RIDER and HN1 datasets, suggesting they may be agnostic to disease site. Useful radiomics features should be selected based on reproducibility. This study identified a set of features that meet this requirement and validated the methodology for evaluating reproducibility between datasets.

Is radiomics a useful addition to magnetic resonance imaging in the preoperative classification of PitNETs?

BACKGROUND: The WHO 2021 introduced the term pituitary neuroendocrine tumours (PitNETs) for pituitary adenomas and incorporated transcription factors for subtyping, prompting the need for fresh diagnostic methods. Current biomarkers struggle to distinguish between high- and low-risk non-functioning PitNETs. We explored if radiomics can enhance preoperative decision-making. METHODS: Pre-treatment magnetic resonance (MR) images of patients who underwent surgery between 2015 and 2019 with available WHO 2021 classification were used. The tumours were manually segmented on the T1w, T1-contrast enhanced, and T2w images using 3D Slicer. One hundred Pyradiomic features were extracted from each MR sequence. Models were built to classify (1) somatotroph and gonadotroph PitNETs and (2) high- and low-risk subtypes of non-functioning PitNETs. Feature were selected independently from the MR sequences and multi-sequence (combining data from more than one MR sequence) using Boruta and Pearson correlation. Support vector machine (SVM), logistic regression (LR), random forest (RF), and multi-layer perceptron (MLP) were the classifiers used. Data imbalance was addressed using the Synthetic Minority Oversampling TEchnique (SMOTE). Performance of the models were evaluated using area under the receiver operating curve (AUC), accuracy, sensitivity, and specificity. RESULTS: A total of 222 PitNET patients (train, n = 149; test, n = 73) were enrolled in this retrospective study. Multi-sequence-based LR model discriminated best between somatotroph and gonadotroph PitNETs, with a test AUC of 0.84, accuracy of 0.74, specificity of 0.81, and sensitivity of 0.70. Multi-sequence-based MLP model perfomed best for the high- and low-risk non-functioning PitNETs, achieving a test AUC of 0.76, accuracy of 0.67, specificity of 0.72, and sensitivity of 0.66. CONCLUSIONS: Utilizing pre-treatment MRI and radiomics holds promise for distinguishing high-risk from low-risk non-functioning PitNETs based on the latest WHO classification. This could assist neurosurgeons in making critical decisions regarding surgery or alternative management strategies for PitNETs after further clinical validation.

Multi-centre radiomics for prediction of recurrence following radical radiotherapy for head and neck cancers: Consequences of feature selection, machine learning classifiers and batch-effect harmonization.

Background and purpose: Radiomics models trained with limited single institution data are often not reproducible and generalisable. We developed radiomics models that predict loco-regional recurrence within two years of radiotherapy with private and public datasets and their combinations, to simulate small and multi-institutional studies and study the responsiveness of the models to feature selection, machine learning algorithms, centre-effect harmonization and increased dataset sizes. Materials and methods: 562 patients histologically confirmed and treated for locally advanced head-and-neck cancer (LA-HNC) from two public and two private datasets; one private dataset exclusively reserved for validation. Clinical contours of primary tumours were not recontoured and were used for Pyradiomics based feature extraction. ComBat harmonization was applied, and LASSO-Logistic Regression (LR) and Support Vector Machine (SVM) models were built. 95% confidence interval (CI) of 1000 bootstrapped area-under-the-Receiver-operating-curves (AUC) provided predictive performance. Responsiveness of the models' performance to the choice of feature selection methods, ComBat harmonization, machine learning classifier, single and pooled data was evaluated. Results: LASSO and SelectKBest selected 14 and 16 features, respectively; three were overlapping. Without ComBat, the LR and SVM models for three institutional data showed AUCs (CI) of 0.513 (0.481-0.559) and 0.632 (0.586-0.665), respectively. Performances following ComBat revealed AUCs of 0.559 (0.536-0.590) and 0.662 (0.606-0.690), respectively. Compared to single cohort AUCs (0.562-0.629), SVM models from pooled data performed significantly better at AUC = 0.680. Conclusions: Multi-institutional retrospective data accentuates the existing variabilities that affect radiomics. Carefully designed prospective, multi-institutional studies and data sharing are necessary for clinically relevant head-and-neck cancer prognostication models.

Utility of Mid-treatment DWI in Selecting Pathological Responders to Neoadjuvant Chemoradiotherapy in Locally Advanced Esophageal Cancer.

PURPOSE: Pathological complete response correlates with better clinical outcomes in locally advanced esophageal cancer (LA-EC). However, there is lack of prognostic markers to identify patients in the current setting of neoadjuvant chemoradiotherapy (NACRT) followed by surgery. This study evaluates the utility of mid-treatment diffusion-weighted imaging (DWI) in identifying pathological responders of NACRT. METHODS: Twenty-four patients with LA-EC on NACRT were prospectively recruited and underwent three MRI (baseline, mid-treatment, end-of-RT) scans. DWI-derived apparent diffusion coefficient (ADC) mean and minimum were used as a surrogate to evaluate the treatment response, and its correlation to pathological response was assessed. RESULTS: Mid-treatment ADC mean was significantly higher among patients with pathological response compared to non-responders (p = 0.011). ADC difference (ΔADC) between baseline and mid-treatment correlated with tumor response (p = 0.007). ADC at other time points did not correlate to pathological response. CONCLUSION: In this study, mid-treatment ADC values show potential to be a surrogate for tumor response in NACRT. However, larger trials are required to establish DW-MRI as a definite biomarker for tumor response.

Comparison of regional lung perfusion response on longitudinal MAA SPECT/CT in lung cancer patients treated with and without functional tissue-avoidance radiation therapy.

OBJECTIVE: The effect of functional lung avoidance planning on radiation dose-dependent changes in regional lung perfusion is unknown. We characterized dose-perfusion response on longitudinal perfusion single photon emission computed tomography (SPECT)/CT in two cohorts of lung cancer patients treated with and without functional lung avoidance techniques. METHODS: The study included 28 primary lung cancer patients: 20 from interventional (NCT02773238) (FLARE-RT) and eight from observational (NCT01982123) (LUNG-RT) clinical trials. FLARE-RT treatment plans included perfused lung dose constraints while LUNG-RT plans adhered to clinical standards. Pre- and 3 month post-treatment macro-aggregated albumin (MAA) SPECT/CT scans were rigidly co-registered to planning four-dimensional CT scans. Tumour-subtracted lung dose was converted to EQD2 and sorted into 5 Gy bins. Mean dose and percent change between pre/post-RT MAA-SPECT uptake (%ΔPERF), normalized to total tumour-subtracted lung uptake, were calculated in each binned dose region. Perfusion frequency histograms of pre/post-RT MAA-SPECT were analyzed. Dose-response data were parameterized by sigmoid logistic functions to estimate maximum perfusion increase (%ΔPERFmaxincrease), maximum perfusion decrease (%ΔPERFmaxdecrease), dose midpoint (Dmid), and dose-response slope (k). RESULTS: Differences in MAA perfusion frequency distribution shape between time points were observed in 11/20 (55%) FLARE-RT and 2/8 (25%) LUNG-RT patients (p < 0.05). FLARE-RT dose response was characterized by >10% perfusion increase in the 0-5 Gy dose bin for 8/20 patients (%ΔPERFmaxincrease = 10-40%), which was not observed in any LUNG-RT patients (p = 0.03). The dose midpoint Dmid at which relative perfusion declined by 50% trended higher in FLARE-RT compared to LUNG-RT cohorts (35 GyEQD2 vs 21 GyEQD2, p = 0.09), while the dose-response slope k was similar between FLARE-RT and LUNG-RT cohorts (3.1-3.2, p = 0.86). CONCLUSION: Functional lung avoidance planning may promote increased post-treatment perfusion in low dose regions for select patients, though inter-patient variability remains high in unbalanced cohorts. These preliminary findings form testable hypotheses that warrant subsequent validation in larger cohorts within randomized or case-matched control investigations. ADVANCES IN KNOWLEDGE: This novel preliminary study reports differences in dose-response relationships between patients receiving functional lung avoidance radiation therapy (FLARE-RT) and those receiving conventionally planned radiation therapy (LUNG-RT). Following further validation and testing of these effects in larger patient populations, individualized estimation of regional lung perfusion dose-response may help refine future risk-adaptive strategies to minimize lung function deficits and toxicity incidence.