Sachin Umrao

Sachin Umrao

Auto-generated profile, based on publicly available information

Sachin Umrao

University of California, San Francisco, USA

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

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

Representative Publications

View All Publications
10 March 2022 in JCO Clinical Cancer Informatics
Verified
Journal Article
Open access
MJMarina Z. Joel
Marina Z. Joel
SUSachin Umrao
Sachin Umrao
ECEnoch Chang
Enoch Chang
RCRachel Choi
Rachel Choi
DYDaniel X. Yang
Daniel X. Yang
SASanjay Aneja
Sanjay Aneja
PURPOSE: Deep learning (DL) models have rapidly become a popular and cost-effective tool for image classification within oncology. A major limitation of DL models is their vulnerability to adversarial images, manipulated input images designed to cause misclassifications by DL models. The purpose of the study is to investigate the robustness of DL models trained on diagnostic images using adversarial images and explore the utility of an iterative adversarial training approach to improve the robustness of DL models against adversarial images. METHODS: We examined the impact of adversarial images on the classification accuracies of DL models trained to classify cancerous lesions across three common oncologic imaging modalities. The computed tomography (CT) model was trained to classify malignant lung nodules. The mammogram model was trained to classify malignant breast lesions. The magnetic resonance imaging (MRI) model was trained to classify brain metastases. RESULTS: Oncologic images showed instability to small pixel-level changes. A pixel-level perturbation of 0.004 (for pixels normalized to the range between 0 and 1) resulted in most oncologic images to be misclassified (CT 25.6%, mammogram 23.9%, and MRI 6.4% accuracy). Adversarial training improved the stability and robustness of DL models trained on oncologic images compared with naive models ([CT 67.7% v 26.9%], mammogram [63.4% vs 27.7%], and MRI [87.2% vs 24.3%]). CONCLUSION: DL models naively trained on oncologic images exhibited dramatic instability to small pixel-level changes resulting in substantial decreases in accuracy. Adversarial training techniques improved the stability and robustness of DL models to such pixel-level changes. Before clinical implementation, adversarial training should be considered to proposed DL models to improve overall performance and safety.

24 November 2016 in IETE Journal of Research
Verified
Journal Article
Closed access
SUSachin Umrao
Sachin Umrao
ARAbhishek Roy
Abhishek Roy
Navrati SaxenaNavrati Saxena
Navrati Saxena
To endure the escalating demand of wireless broadband applications such as high-definition videos, live video streaming in 4G/beyond 4G cellular networks (5G), Device-to-Device (D2D) communication along with D2D caching can serve as a candidate paradigm to enhance spectrum efficiency and Quality of Service (QoS). D2D users can establish a direct connection without routing through evolved NodeB (eNB) and legacy network, by reusing the cellular spectrum, which helps in boosting spectral efficiency. Furthermore, the rapid increase in storage capacity of smart devices can positively drive the network's QoS by caching locally popular videos and disseminating them via D2D communication. With this motivation, we propose a video dissemination architecture with MIX caching, which includes the fusion of evolved packet core (EPC) caching, eNB caching, and additional Friend Node (FN). This paper presents a novel approach, specifically caching popular videos in smartphone (Friend Node) and disseminating by D2D communication. By placing popular data close to users, this approach facilitates users to download data from FN in proximity enabling significant delay reduction. The proposed work advocates fusion of FN cache in operator controlled long-term evolution-advanced system to improve network performance in terms of delay and jitter. Our simulation result shows that FN caching reduces delay up to 75% while MIX caching reduces delay ranging from 15% to 57%. We also analyze downloading time of various video files under D2D, eNB, EPC, and MIX caching.

16 June 2026 in HAND
Verified
Journal Article
Open access
MKMitchell R. Koss
Mitchell R. Koss
LWLiam Wong
Liam Wong
TPThomas A. Peterson
Thomas A. Peterson
SUSachin Umrao
Sachin Umrao
GLGopal R. Lalchandani
Gopal R. Lalchandani
Background: Delay in surgical fixation of distal radius fractures (DRFs) beyond 2 weeks is associated with worse patient-reported outcomes. Although socioeconomic disadvantage affects many orthopedic outcomes, its association with delayed DRF fixation is unclear. We evaluated whether neighborhood disadvantage is associated with surgical delay beyond 14 days. Methods: This retrospective study used deidentified electronic health records from multiple academic health systems to identify patients undergoing operative DRF fixation from 2012 to 2024. Multivariable logistic regression identified socioeconomic and demographic factors associated with delay beyond 14 days. Covariates included age, sex, area deprivation index (ADI), race, ethnicity, language, Medicare insurance, and Charlson Comorbidity Index (CCI). Results: The cohort included 6639 patients (median age 62 years; 64.7% female; 64.8% white; 90.4% English-speaking; 16.1% Medicare). Median time from diagnosis to surgery was 8 days (IQR 5-13), with 25.3% delayed beyond 14 days. Higher ADI (quartile 3: OR 1.59, 95% CI 1.24 to 1.81; quartile 4: OR 1.47, 95% CI 1.20 to 1.80), Spanish-speaking status (OR 1.48, 95% CI 1.08 to 2.02), higher CCI (OR 1.10, 95% CI 1.06 to 1.13), and Medicare insurance (OR 1.25, 95% CI 1.00 to 1.55) were independently associated with delay. Conclusions: Socioeconomic disadvantage, limited English proficiency, comorbidity burden, and Medicare insurance were independently associated with delayed DRF fixation. These findings highlight existing disparities in timely orthopedic care of DRFs and support targeted interventions to improve access.