scholarly journals The Challenge of Choosing the Best Classification Method in Radiomic Analyses: Recommendations and Applications to Lung Cancer CT Images

Cancers ◽  
2021 ◽  
Vol 13 (12) ◽  
pp. 3088
Author(s):  
Federica Corso ◽  
Giulia Tini ◽  
Giuliana Lo Presti ◽  
Noemi Garau ◽  
Simone Pietro De Angelis ◽  
...  
Keyword(s):  
Lung Cancer ◽  
Sample Size ◽  
Sensitivity And Specificity ◽  
Simulated Data ◽  
Lymph Node Status ◽  
Imaging Features ◽  
Association Strength ◽  
Operating Characteristics ◽  
Computed Tomography Images ◽  
Extreme Gradient Boosting

Radiomics uses high-dimensional sets of imaging features to predict biological characteristics of tumors and clinical outcomes. The choice of the algorithm used to analyze radiomic features and perform predictions has a high impact on the results, thus the identification of adequate machine learning methods for radiomic applications is crucial. In this study we aim to identify suitable approaches of analysis for radiomic-based binary predictions, according to sample size, outcome balancing and the features–outcome association strength. Simulated data were obtained reproducing the correlation structure among 168 radiomic features extracted from Computed Tomography images of 270 Non-Small-Cell Lung Cancer (NSCLC) patients and the associated to lymph node status. Performances of six classifiers combined with six feature selection (FS) methods were assessed on the simulated data using AUC (Area Under the Receiver Operating Characteristics Curves), sensitivity, and specificity. For all the FS methods and regardless of the association strength, the tree-based classifiers Random Forest and Extreme Gradient Boosting obtained good performances (AUC ≥ 0.73), showing the best trade-off between sensitivity and specificity. On small samples, performances were generally lower than in large–medium samples and with larger variations. FS methods generally did not improve performances. Thus, in radiomic studies, we suggest evaluating the choice of FS and classifiers, considering specific sample size, balancing, and association strength.

scholarly journals Accuracy of pre-hospital triage tools for major trauma: a systematic review with meta-analysis and net clinical benefit

2021 ◽  
Vol 16 (1) ◽  
Author(s):  
Silvia Gianola ◽  
Greta Castellini ◽  
Annalisa Biffi ◽  
Gloria Porcu ◽  
Andrea Fabbri ◽  
...  
Keyword(s):  
Systematic Review ◽  
Quality Assessment ◽  
Sample Size ◽  
Sensitivity And Specificity ◽  
Clinical Benefit ◽  
Major Trauma ◽  
Trauma System ◽  
Operating Characteristics ◽  
Large Variability ◽  
Net Clinical Benefit

Abstract Background We conducted a systematic review to evaluate and compare the accuracy of pre-hospital triage tools for major trauma in the context of the development of the Italian National Institute of Health guidelines on major trauma integrated management. Methods PubMed, Embase, and CENTRAL were searched up to November 2019 for studies investigating pre-hospital triage tools. The ROC (receiver operating characteristics) curve and net clinical benefit for all selected triage tools were performed. Quality assessment was performed using the Quality Assessment of Diagnostic Accuracy Studies–2. Certainty of the evidence was judged with the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. Results We found 15 observational studies of 13 triage tools for adults and 11 for children. In adults, according to the ROC curve and the net clinical benefit, the most reliable tool was the Northern French Alps Trauma System (TRENAU), adopting injury severity score (ISS) > 15 as reference (sensitivity (Sn), 0.92; specificity (Sp), 0.41; 1 study; sample size, 2572; high certainty of the evidence). When mortality as reference was considered, the pre-hospital triage tool with the best net clinical benefit trajectory was the New Trauma Score (NTS) < 18 (Sn, 0.82; Sp, 0.86; 1 study; sample size, 1001; moderate certainty of the evidence). In children, high variability among all triage tools for sensitivity and specificity was found. Conclusion Sensitivity and specificity varied across all available pre-hospital trauma triage tools. TRENAU and NTS are the best accurate triage tools for adults, whereas in the pediatric area a large variability prevents any firm conclusion.

Constructing and model-fitting receiver operator characteristics using continuous data.

2018 ◽  
Author(s):  
Josephine Ann Urquhart ◽  
Akira O'Connor
Keyword(s):  
Signal Detection ◽  
Least Squares Method ◽  
Model Fitting ◽  
Simulated Data ◽  
Continuous Variable ◽  
Continuous Data ◽  
Model Parameters ◽  
Operating Characteristics ◽  
Unequal Variance ◽  
Discrete Discrimination

Receiver operating characteristics (ROCs) are plots which provide a visual summary of a classifier’s decision response accuracy at varying discrimination thresholds. Typical practice, particularly within psychological studies, involves plotting an ROC from a limited number of discrete thresholds before fitting signal detection parameters to the plot. We propose that additional insight into decision-making could be gained through increasing ROC resolution, using trial-by-trial measurements derived from a continuous variable, in place of discrete discrimination thresholds. Such continuous ROCs are not yet routinely used in behavioural research, which we attribute to issues of practicality (i.e. the difficulty of applying standard ROC model-fitting methodologies to continuous data). Consequently, the purpose of the current article is to provide a documented method of fitting signal detection parameters to continuous ROCs. This method reliably produces model fits equivalent to the unequal variance least squares method of model-fitting (Yonelinas et al., 1998), irrespective of the number of data points used in ROC construction. We present the suggested method in three main stages: I) building continuous ROCs, II) model-fitting to continuous ROCs and III) extracting model parameters from continuous ROCs. Throughout the article, procedures are demonstrated in Microsoft Excel, using an example continuous variable: reaction time, taken from a single-item recognition memory. Supplementary MATLAB code used for automating our procedures is also presented in Appendix B, with a validation of the procedure using simulated data shown in Appendix C.

scholarly journals Effects of deep inspiration breath hold on prone photon or proton irradiation of breast and regional lymph nodes

2021 ◽  
Vol 11 (1) ◽  
Author(s):  
Bruno Speleers ◽  
Max Schoepen ◽  
Francesca Belosi ◽  
Vincent Vakaet ◽  
Wilfried De Neve ◽  
...  
Keyword(s):  
Lung Cancer ◽  
At Risk ◽  
Lymph Node ◽  
Organs At Risk ◽  
Lymph Node Status ◽  
Breath Hold ◽  
Deep Inspiration ◽  
Proton Radiotherapy ◽  
Regional Lymph Nodes ◽  
Deep Inspiration Breath Hold

AbstractWe report on a comparative dosimetrical study between deep inspiration breath hold (DIBH) and shallow breathing (SB) in prone crawl position for photon and proton radiotherapy of whole breast (WB) and locoregional lymph node regions, including the internal mammary chain (LN_MI). We investigate the dosimetrical effects of DIBH in prone crawl position on organs-at-risk for both photon and proton plans. For each modality, we further estimate the effects of lung and heart doses on the mortality risks of different risk profiles of patients. Thirty-one patients with invasive carcinoma of the left breast and pathologically confirmed positive lymph node status were included in this study. DIBH significantly decreased dose to heart for photon and proton radiotherapy. DIBH also decreased lung doses for photons, while increased lung doses were observed using protons because the retracting heart is displaced by low-density lung tissue. For other organs-at-risk, DIBH resulted in significant dose reductions using photons while minor differences in dose deposition between DIBH and SB were observed using protons. In patients with high risks for cardiac and lung cancer mortality, average thirty-year mortality rates from radiotherapy-related cardiac injury and lung cancer were estimated at 3.12% (photon DIBH), 4.03% (photon SB), 1.80% (proton DIBH) and 1.66% (proton SB). The radiation-related mortality risk could not outweigh the ~ 8% disease-specific survival benefit of WB + LN_MI radiotherapy in any of the assessed treatments.

Validation of Michigan risk score and D-dimer to predict peripherally inserted central catheter-related thrombosis: A study of 206,132 catheter days

2021 ◽  
pp. 112972982110087
Author(s):  
Junren Kang ◽  
Wenyan Sun ◽  
Hailong Li ◽  
En ling Ma ◽  
Wei Chen
Keyword(s):  
Sensitivity And Specificity ◽  
Risk Score ◽  
Primary Objective ◽  
Class Iii ◽  
Operating Characteristics ◽  
Central Venous ◽  
Sensitivity Specificity ◽  
Catheter Related Thrombosis ◽  
D Dimer ◽  
Independent Cohort

Background: The Michigan Risk Score (MRS) was the only predicted score for peripherally inserted central venous catheters (PICC) associated upper extremity venous thrombosis (UEVT). Age-adjusted D-dimer increased the efficiency for UEVT. There were no external validations in an independent cohort. Method: A retrospective study of adult patients with PICC insertion was performed. The primary objective was to evaluate the performance of the MRS and age-adjusted D-dimer in estimating risk of PICC-related symptomatic UEVT. The sensitivity, specificity and areas under the receiver operating characteristics (ROC) of MRS and age-adjusted D-dimer were calculated. Results: Two thousand one hundred sixty-three patients were included for a total of 206,132 catheter days. Fifty-six (2.6%) developed PICC-UEVT. The incidences of PICC-UEVT were 4.9% for class I, 7.5% for class II, 2.2% for class III, 0% for class IV of MRS ( p = 0.011). The incidences of PICC-UEVT were 4.5% for D-dimer above the age-adjusted threshold and 1.5% for below the threshold ( p = 0.001). The areas under ROC of MRS and age-adjusted D-dimer were 0.405 (95% confidence interval (CI) 0.303–0.508) and 0.639 (95% CI 0.547–0.731). The sensitivity and specificity of MRS were 0.82 (95% CI, 0.69–0.91), 0.09 (95% CI, 0.08–0.11), respectively. The sensitivity and specificity of age-adjusted D-dimer were 0.64 (95% CI, 0.46–0.79) and 0.64 (95% CI, 0.61–0.66), respectively. Conclusions: MRS and age-adjusted D-dimer have low accuracy to predict PICC-UEVT. Further studies are needed.

Alternative models and randomization techniques for Bayesian response-adaptive randomization with binary outcomes

2021 ◽  
pp. 174077452110101
Author(s):  
Jennifer Proper ◽  
John Connett ◽  
Thomas Murray
Keyword(s):  
Logistic Regression ◽  
Sample Size ◽  
Error Rate ◽  
Adaptive Design ◽  
Type I Error ◽  
Probability Model ◽  
Binary Outcomes ◽  
Type I ◽  
Operating Characteristics ◽  
Type I Error Rate

Background: Bayesian response-adaptive designs, which data adaptively alter the allocation ratio in favor of the better performing treatment, are often criticized for engendering a non-trivial probability of a subject imbalance in favor of the inferior treatment, inflating type I error rate, and increasing sample size requirements. The implementation of these designs using the Thompson sampling methods has generally assumed a simple beta-binomial probability model in the literature; however, the effect of these choices on the resulting design operating characteristics relative to other reasonable alternatives has not been fully examined. Motivated by the Advanced R2 Eperfusion STrategies for Refractory Cardiac Arrest trial, we posit that a logistic probability model coupled with an urn or permuted block randomization method will alleviate some of the practical limitations engendered by the conventional implementation of a two-arm Bayesian response-adaptive design with binary outcomes. In this article, we discuss up to what extent this solution works and when it does not. Methods: A computer simulation study was performed to evaluate the relative merits of a Bayesian response-adaptive design for the Advanced R2 Eperfusion STrategies for Refractory Cardiac Arrest trial using the Thompson sampling methods based on a logistic regression probability model coupled with either an urn or permuted block randomization method that limits deviations from the evolving target allocation ratio. The different implementations of the response-adaptive design were evaluated for type I error rate control across various null response rates and power, among other performance metrics. Results: The logistic regression probability model engenders smaller average sample sizes with similar power, better control over type I error rate, and more favorable treatment arm sample size distributions than the conventional beta-binomial probability model, and designs using the alternative randomization methods have a negligible chance of a sample size imbalance in the wrong direction. Conclusion: Pairing the logistic regression probability model with either of the alternative randomization methods results in a much improved response-adaptive design in regard to important operating characteristics, including type I error rate control and the risk of a sample size imbalance in favor of the inferior treatment.

Independent validation of simbiosys tumorscope to predict response to neoadjuvant chemotherapy (NACT) in early breast cancer (EBC).

2021 ◽  
Vol 39 (15_suppl) ◽  
pp. 582-582
Author(s):  
Gong He ◽  
Frederick Howard ◽  
Tushar Pandey ◽  
Hiroyuki Abe ◽  
Rita Nanda
Keyword(s):  
Breast Cancer ◽  
Sensitivity And Specificity ◽  
Cancer Biology ◽  
Tumor Volume ◽  
Performance Metrics ◽  
External Validation ◽  
Pathologic Complete Response ◽  
Disease Free Survival ◽  
Biophysical Model ◽  
Lymph Node Status

582 Background: Despite substantial advances in the understanding of breast cancer biology, the decision to use NACT for EBC is based on tumor size, lymph node status, and subtype. Even with aggressive therapy, the majority of women will not achieve a pathologic complete response (pCR). Investigational treatment regimens, including immunotherapy, can increase pCR rates, but are associated with irreversible immune-related toxicities. Being able to accurately predict pCR could identify candidates for intensification or de-escalation of NACT, allowing for personalized medicine. SimBioSys TumorScope (TS) is a biophysical model that utilizes baseline MRI, receptor status, and planned treatment regimen to simulate response to NACT over time. TS has demonstrated accurate prediction of pCR in prior studies. Here, we describe an independent external validation of TS. Methods: We conducted a retrospective study of University of Chicago patients (pts) who received NACT for EBC from Jan 2010 - March 2020. Pts must have had a pretreatment breast MRI. Tumors were analyzed using TS by investigators who were blinded to response data. TS predicted pCR was predefined as a residual tumor volume < 0.01 cm3 or a 99.9% or greater reduction in tumor volume. Performance metrics of TS were calculated. Results: 144 tumors from 141 pts were analyzed. Average age was 52 yrs; 65% had stage II and 19% had stage III disease. Sensitivity and specificity of TS for predicting pCR were 90.4% and 92.4%, respectively. Of the 7 patients who were predicted to achieve a pCR but did not, 5 had a tumor cellularity < 5%. With a median follow-up of 4.7 yrs, the 4-yr distant disease free survival (DDFS) was 100% for patients predicted to achieve pCR, versus 81.5% for those predicted to have residual disease. Results were generally robust for all subgroups analyzed (Table). Conclusions: TS accurately predicts pCR and DDFS from baseline MRI and clinicopathologic data. Given the high sensitivity and specificity of this assay across breast cancer subtypes, TS can be used to aid in escalation/de-escalation strategies for EBC.[Table: see text]

Sign in / Sign up

Export Citation Format

Share Document