Stella K. Kang

Stella K. Kang

Auto-generated profile, based on publicly available information

Stella K. Kang

Department of Radiology, Columbia University Irving Medical Center, New York, NY

Research Focus & Metrics

Academic Career Length

18Years (2009-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 September 2022 in Science Translational Medicine
Verified
Journal Article
Open access
JWJan Witowski
Jan Witowski
LHLaura Heacock
Laura Heacock
BRBeatriu Reig
Beatriu Reig
SKStella K. Kang
Stella K. Kang
ALAlana Lewin
Alana Lewin
KGKrzysztof J. Geras
Krzysztof J. Geras
Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has a high sensitivity in detecting breast cancer but often leads to unnecessary biopsies and patient workup. We used a deep learning (DL) system to improve the overall accuracy of breast cancer diagnosis and personalize management of patients undergoing DCE-MRI. On the internal test set ( n = 3936 exams), our system achieved an area under the receiver operating characteristic curve (AUROC) of 0.92 (95% CI: 0.92 to 0.93). In a retrospective reader study, there was no statistically significant difference ( P = 0.19) between five board-certified breast radiologists and the DL system (mean ΔAUROC, +0.04 in favor of the DL system). Radiologists’ performance improved when their predictions were averaged with DL’s predictions [mean ΔAUPRC (area under the precision-recall curve), +0.07]. We demonstrated the generalizability of the DL system using multiple datasets from Poland and the United States. An additional reader study on a Polish dataset showed that the DL system was as robust to distribution shift as radiologists. In subgroup analysis, we observed consistent results across different cancer subtypes and patient demographics. Using decision curve analysis, we showed that the DL system can reduce unnecessary biopsies in the range of clinically relevant risk thresholds. This would lead to avoiding biopsies yielding benign results in up to 20% of all patients with BI-RADS category 4 lesions. Last, we performed an error analysis, investigating situations where DL predictions were mostly incorrect. This exploratory work creates a foundation for deployment and prospective analysis of DL-based models for breast MRI.

7 January 2017 in Abdominal Radiology
Verified
Journal Article
Closed access
PDF file
NWNicole Wake
Nicole Wake
TRTemitope Rude
Temitope Rude
SKStella K. Kang
Stella K. Kang
MSMichael D. Stifelman
Michael D. Stifelman
JBJames F. Borin
James F. Borin
HCHersh Chandarana
Hersh Chandarana

1 June 2014 in American Journal of Roentgenology
Verified
Journal Article
Closed access
SKStella K. Kang
Stella K. Kang
WHWilliam C. Huang
William C. Huang
PPPari V. Pandharipande
Pari V. Pandharipande
HCHersh Chandarana
Hersh Chandarana
OBJECTIVE. Solid renal masses are most often incidentally detected at imaging as small (≤ 4 cm) localized lesions. These lesions comprise a wide spectrum of benign and malignant histologic subtypes...

1 December 2012 in Investigative Radiology
Verified
Journal Article
Closed access
HCHersh Chandarana
Hersh Chandarana
SKStella K. Kang
Stella K. Kang
SWSamson Wong
Samson Wong
HRHenry Rusinek
Henry Rusinek
JZJeff Lei Zhang
Jeff Lei Zhang
ESEric Sigmund
Eric Sigmund
The aim of this study was to use intravoxel incoherent motion diffusion-weighted imaging to discriminate subtypes of renal neoplasms and to assess agreement between intravoxel incoherent motion (perfusion fraction, fp) and dynamic contrast-enhanced magnetic resonance imaging (MRI) metrics of tumor vascularity. In this Health Insurance Portability and Accountability Act-compliant, institutional review board-approved prospective study, 26 patients were imaged at 1.5-T MRI using dynamic contrast-enhanced MRI with high temporal resolution and diffusion-weighted imaging using 8 b values (range, 0-800 s/mm). Perfusion fraction (fp), tissue diffusivity (Dt), and pseudodiffusivity (Dp) were calculated using biexponential fitting of the diffusion data. Apparent diffusion coefficient (ADC) was calculated with monoexponential fit using 3 b values of 0, 400, and 800 s/mm. Dynamic contrast-enhanced data were processed with a semiquantitative method to generate model-free parameter cumulative initial area under the curve of gadolinium concentration at 60 seconds (CIAUC60). Perfusion fraction, Dt, Dp, ADC, and CIAUC60 were compared between different subtypes of renal lesions. Perfusion fraction was correlated with CIAUC60. We examined 14 clear cell, 4 papillary, 5 chromophobe, and 3 cystic renal cell carcinomas (RCCs). Although fp had higher accuracy (area under the curve, 0.74) for a diagnosis of clear cell RCC compared with Dt or ADC, the combination of fp and Dt had the highest accuracy (area under the curve, 0.78). The combination of fp and Dt diagnosed papillary RCC and cystic RCC with 100% accuracy, and clear cell RCC and chromophobe RCC, with 86.5% accuracy. There was significant strong correlation between fp and CIAUC60 (r = 0.82; P < 0.001). Intravoxel incoherent motion parameters fp and Dt can discriminate renal tumor subtypes. Perfusion fraction demonstrates good correlation with CIAUC60 and can assess degree of tumor vascularity without the use of exogenous contrast agent.

1 January 2016 in Radiology
Verified
Journal Article
Closed access
AKAndrea S. Kierans
Andrea S. Kierans
SKStella K. Kang
Stella K. Kang
ARAndrew B. Rosenkrantz
Andrew B. Rosenkrantz
Results of our meta-analysis suggest that dynamic contrast-enhanced MR imaging has excellent specificity and moderate, although highly variable, sensitivity in the detection of small hepatocellular carcinoma that measures up to 2 cm. PurposeTo determine the performance of dynamic contrast material–enhanced magnetic resonance (MR) imaging in the diagnosis of small (≤2-cm) hepatocellular carcinoma (HCC) and to identify factors that influence this performance.Materials and MethodsMedline and Embase databases were searched for studies performed from January 2000 to March 2014 in which the performance of MR imaging was reported for the detection of HCC up to 2 cm on either a lesion- or patient-based level, with sufficient data to construct 2 × 2 contingency tables. Diagnostic performance was quantitatively pooled for all studies by using a bivariate random-effects model with exploration involving subgroup analysis, meta-regression, and determination of study heterogeneity.ResultsTwenty-two studies with 1387 small HCC lesions in 1908 patients met inclusion criteria. Heterogeneity was higher for sensitivity (range, 30%–99%) than specificity (range, 61%–100%). Overall sensitivity was 78% (95% confidence interval [CI]: 68%, 85%; I2 = 89%), and...

3 June 2020 in Lab on a Chip
Verified
Journal Article
Open access
PDF file
MMMichael P. McRae
Michael P. McRae
GSGlennon W. Simmons
Glennon W. Simmons
NCNicolaos Christodoulides
Nicolaos Christodoulides
ZLZhibing Lu
Zhibing Lu
SKStella K. Kang
Stella K. Kang
JMJohn Mcdevitt
John Mcdevitt
SARS-CoV-2 is the virus that causes coronavirus disease (COVID-19) which has reached pandemic levels resulting in significant morbidity and mortality affecting every inhabited continent. The large number of patients requiring intensive care threatens to overwhelm healthcare systems globally. Likewise, there is a compelling need for a COVID-19 disease severity test to prioritize care and resources for patients at elevated risk of mortality. Here, an integrated point-of-care COVID-19 Severity Score and clinical decision support system is presented using biomarker measurements of C-reactive protein (CRP), N-terminus pro B type natriuretic peptide (NT-proBNP), myoglobin (MYO), D-dimer, procalcitonin (PCT), creatine kinase-myocardial band (CK-MB), and cardiac troponin I (cTnI). The COVID-19 Severity Score combines multiplex biomarker measurements and risk factors in a statistical learning algorithm to predict mortality. The COVID-19 Severity Score was trained and evaluated using data from 160 hospitalized COVID-19 patients from Wuhan, China. Our analysis finds that COVID-19 Severity Scores were significantly higher for the group that died versus the group that was discharged with median (interquartile range) scores of 59 (40–83) and 9 (6–17), respectively, and area under the curve of 0.94 (95% CI 0.89–0.99). Although this analysis represents patients with cardiac comorbidities (hypertension), the inclusion of biomarkers from other pathophysiologies implicated in COVID-19 (e.g., D-dimer for thrombotic events, CRP for infection or inflammation, and PCT for bacterial co-infection and sepsis) may improve future predictions for a more general population. These promising initial models pave the way for a point-of-care COVID-19 Severity Score system to impact patient care after further validation with externally collected clinical data. Clinical decision support tools for COVID-19 have strong potential to empower healthcare providers to save lives by prioritizing critical care in patients at high risk for adverse outcomes.