Search

Introduction: Diffusion weighted MRI (DWI) has emerged as a promising adjunct to reduce unnecessary biopsies prompted by breast MRI through use of apparent diffusion coefficient (ADC) measures. The purpose of this study was to investigate the effects of different lesion ADC measurement approaches and ADC cutoffs on the diagnostic performance of breast DWI in a high-risk MRI screening cohort to identify the optimal approach for clinical incorporation. Methods: Consecutive screening breast MRI examinations (August 2014–Dec 2018) that prompted a biopsy for a suspicious breast lesion (BI-RADS 4 or 5) were retrospectively evaluated. On DWI, ADC (b=0/100/600/800s/mm2) measures were calculated with three different techniques for defining lesion region-of-interest (ROI; single slice('2D'), whole volume('3D') and lowest ADC region('hotspot')). An optimal data-derived ADC cutoff for each technique was retrospectively identified to reduce benign biopsies while avoiding any false negatives, inherently producing cutoffs with 100% sensitivity in this particular cohort. Further, diagnostic performance of these measures was validated using two prespecified ADC cutoffs: 1.53x10-3mm2/s from the ECOG-ACRIN A6702 trial and 1.30x10-3mm2/s from the international EUSOBI group. Diagnostic performance was compared between ADC maps generated with 2(0/800s/mm2) and 4(0/100/600/800s/mm2) b-values. Benign biopsy reduction rate was calculated (number of benign lesions with ADC >cutoff)/(total number of benign lesions). Results: 137 suspicious lesions (in 121 women, median age 44 years [range, 20-75yrs]) were detected on contrast-enhanced screening breast MRI and recommended for biopsy. Of those, 30(21.9%) were malignant and 107(78.1%) were benign. Hotspot ADC measures were significantly lower (p<0.001) than ADCs from both 2D and 3D ROI techniques. Applying the optimal data-derived ADC cutoffs resulted in comparable reduction in benign biopsies across ROI techniques (range:16.8% -17.8%). Applying the prespecified A6702 and EUSOBI cutoffs resulted in benign biopsy reduction rates of 11.2-19.6%(with 90.0-100% sensitivity) and 36.4-51.4%(with 70.0-83.3% sensitivity), respectively, across ROI techniques. ADC measures and benign biopsy reduction rates were similar when calculated with only 2 b-values (0,800 s/mm2) versus all 4 b-values. Discussion: Our findings demonstrate that with appropriate ADC thresholds, comparable reduction in benign biopsies can be achieved using lesion ADC measurements computed from a variety of approaches. Choice of ADC cutoff depends on ROI approach and preferred performance tradeoffs (biopsy reduction vs sensitivity). [ABSTRACT FROM AUTHOR]

Background Patients with mycosis fungoides (MF)/Sézary Syndrome (SS) can experience impacted health-related quality of life (HRQoL). Objectives To validate the CTCL-S, a novel subscale of the Functional Assessment of Cancer Therapy–General (FACT-G), in patients with MF/SS. Methods Qualitative interviews were conducted with expert clinicians and patients with MF/SS. Thematic analysis identified the most common concerns, and 19 items were selected. Patients with MF/SS were recruited from a single centre. FACT-G, CTCL-S (collectively 'FACT-CTCL'), Skindex-29 and Visual Analogue Scale–Pruritis (VAS-itch) were administered. A subset repeated FACT-CTCL and VAS-itch after ≈2 weeks. Patient demographics and clinical characteristics were obtained via review of the electronic medical records. Psychometric properties were assessed. Internal consistency was estimated using Cronbach's α. Convergent and discriminant validity were assessed by comparing CTCL-S with disease stage, age, VAS-itch, FACT-G and Skindex-29. Exploratory factor analysis (EFA) was used to preliminarily assess CTCL-S dimensionality. Test–retest repeatability was summarized using intraclass correlation coefficient (ICC), within-subject standard deviation and within-subject coefficient of variation. Results Seventy-two patients completed the initial survey, and 35 repeated the FACT-CTCL and VAS-itch after ≈2 weeks. Two-thirds were men; most were White (78%). The majority (85%) had MF, 15% had SS and 75% early (stage IA–IIA) and 25% advanced (≥ stage IIB) disease. Preliminary EFA found a single predominant factor, supporting a hypothesis of unidimensionality of the CTCL-S. Internal consistency of the CTCL-S was high, with α = 0.95 [95% confidence interval (CI) 0.93–0.96]. There was no significant change in CTCL-S average test–retest scores [ICC 0.93 (P = 0.63)]. CTCL-S was significantly lower in advanced vs. early-stage disease [median (interquartile range) 34 (26–48) vs. 59 (44–68), P < 0.001] and strongly correlated with VAS-itch [Spearman's r (rs) –0.70, 95% CI –0.81 to –0.55], FACT-G (rs 0.77, 95% CI 0.65–0.85) and Skindex-29 (rs –0.90, 95% CI –0.94 to –0.84), supporting convergent validity. CTCL-S scores had little correlation with age (rs 0.19, 95% CI –0.05 to 0.41, P = 0.12), supporting discriminant validity. Conclusions The FACT-CTCL is a disease-specific instrument for assessing HRQoL with high reproducibility and good performance in a cohort of patients with MF/SS. [ABSTRACT FROM AUTHOR]

Artificial Intelligence (AI) algorithms showcase the potential to steer a paradigm shift in clinical medicine, especially medical imaging. Concerns associated with model generalizability and biases necessitate rigorous external validation of AI algorithms prior to their adoption into clinical workflows. To address the barriers associated with patient privacy, intellectual property, and diverse model requirements, we introduce ClinValAI, a framework for establishing robust cloud-based infrastructures to clinically validate AI algorithms in medical imaging. By featuring dedicated workflows for data ingestion, algorithm scoring, and output processing, we propose an easily customizable method to assess AI models and investigate biases. Our novel orchestration mechanism facilitates utilizing the complete potential of the cloud computing environment. ClinValAI's input auditing and standardization mechanisms ensure that inputs consistent with model prerequisites are provided to the algorithm for a streamlined validation. The scoring workflow comprises multiple steps to facilitate consistent inferencing and systematic troubleshooting. The output processing workflow helps identify and analyze samples with missing results and aggregates final outputs for downstream analysis. We demonstrate the usability of our work by evaluating a state-of-the-art breast cancer risk prediction algorithm on a large and diverse dataset of 2D screening mammograms. We perform comprehensive statistical analysis to study model calibration and evaluate performance on important factors, including breast density, age, and race, to identify latent biases. ClinValAI provides a holistic framework to validate medical imaging models and has the potential to advance the development of generalizable AI models in clinical medicine and promote health equity.