About

Saurabh Gupta is an Assistant Professor at the Electrical & Computer Engineering department within the Siebel School of Computing and Data Science at the University of Illinois Urbana-Champaign. His research areas include Artificial Intelligence, with recent courses taught focusing on Deep Learning for Computer Vision, Learning-Based Robotics, and Robot Learning. His work involves building predictive models to improve robot navigation and object interaction, contributing to advancements in robotics and intelligent systems. He has earned recognition through the NSF CAREER award for his efforts in developing innovative approaches in these fields.

Research topics

• Computer Science
• Data Mining

Selected publications

• Edoxaban versus Apixaban Outcomes Differences in 8,444 Patients with Atrial Fibrillation from Italy: A Real-World Use Comparison

  TH Open · 2026-03-23

  articleOpen access

  ABSTRACT Based on efficacy and safety data from randomized controlled trials (RCTs) and real-world studies, direct oral anticoagulants are recommended for thromboembolism prevention in patients with atrial fibrillation (AF). However, no RCTs have compared edoxaban with apixaban. This study aimed to retrospectively compare clinical outcomes for patients with AF in Italy (overall and by age group) who received edoxaban versus apixaban. Adult patients with AF who newly initiated an edoxaban or apixaban prescription between January 2016 and December 2021 were identified from the Italian IQVIA® Longitudinal Patient Database. Patient characteristics were summarized. Propensity score matching was used to balance baseline characteristics between the edoxaban and apixaban groups. Clinical outcomes of effectiveness (ischemic stroke [IS] or systemic embolism [SE]) and safety (any major bleeding [MB]) were compared. Incidence rates per 100 person-years and hazard ratios (HRs) with 95% CIs were computed. Among 8,444 identified patients, 37.8% (n = 3,188) were prescribed edoxaban and 62.2% (n = 5,256) were prescribed apixaban. After matching, patient characteristics were similar between cohorts. The post-matching risk (HR, 95% CI) of IS/SE was significantly lower for edoxaban versus apixaban in the overall population (0.78, 0.61–0.99; p = 0.04) and in patients aged ≥80 years (0.61, 0.44–0.86; p < 0.01), with a similar risk for MB for edoxaban versus apixaban. No significant differences were observed between edoxaban and apixaban among patients aged <80 years (all p > 0.05). IS/SE risk was significantly lower for edoxaban versus apixaban, without an increased MB risk among patients with AF overall and those aged ≥80 years.

  PublisherOA PDFDOI

• Real-time banana harvest readiness prediction using mobile SE-enhanced YOLO classification

  Journal of Agricultural Engineering · 2026-01-28

  articleOpen access

  A digital banana harvesting solution was developed to improve the speed and consistency of banana harvesting by integrating real-time bunch detection with harvest-readiness classification into a mobile decision support system used directly in the field. The banana bunch detection module utilizes a You Only Look Once (YOLO) model trained on a custom dataset collected under real plantation conditions, enabling consistent performance across varied environments. Specifically, a YOLOv12n detector was used for banana bunch detection, achieving 93% AP50-test with an inference latency of 5.1 ms per image, making it suitable for mobile deployment in plantation environments. For the readiness of harvesting prediction, a second model was developed, based on a squeeze-and-excitation YOLO classifier, using annotated images gathered with guidance from harvesting experts. In this work, this SE-enhanced YOLO classifier is used as a lightweight, task-specific YOLO classification backbone for the binary “cut” vs “keep” decision, and this harvest-readiness classifier achieved 94% accuracy with an inference time of 2.8 ms per image. Then, an application was built using Flutter and Dart, which uses intuitive interfaces for both field operators and administrators, and includes integrated feedback mechanisms to collect user input and support continuous model refinement. Field testing across diverse lighting and environmental conditions, as well as usability assessments with expert harvesters and administrative staff, demonstrated reliable performance with potential to contribute to faster decision-making and reduced manual labour.

  PublisherOA PDFDOI

• Technology-Facilitated Abuse In Domestic Relationships: ‘A Growing Threat To Women

  Lex localis - Journal of Local Self-Government · 2025-10-03

  articleOpen access

  The trend to use digital and communication technology for controlling and abusing a life partner in a domestic relationship is rampant. Promotion of the ideas of patriarchy and gender stereotyping leads to violence against women in a family setup. Domestic violence against women is a serious issue worldwide. It is globally prevailing, deeply rooted and leaves serious impacts on women's health and well-being. Domestic violence, in its broadest sense, refers to abuse – physical, emotional, sexual, or financial – of one living partner by the other, often living in the same household. Abuse of networking platforms, applications meant for surveillance, spyware and other tracking devices are a few examples of digital abuse. This type of abuse of technological means can be referred to as ‘technology-facilitated’ abuse.

  PublisherOA PDFDOI

• Antimicrobial Spices: Use in Antimicrobial Packaging

  Elsevier eBooks · 2025-01-01 · 1 citations

  book-chapterSenior author
  PublisherDOI

• Treatment Patterns of Goserelin 3.6 mg Once Every 4 Weeks and 10.8 mg Once Every 12 Weeks in Women With Breast Cancer: A Real-World Analysis of Patients in the United States

  JCO Oncology Practice · 2025-05-21

  articleOpen access

  PURPOSE: Goserelin is a gonadotropin-releasing hormone agonist for ovarian function suppression in the treatment of pre- and perimenopausal patients with breast cancer and for the preservation of ovarian function during chemotherapy. Goserelin is available in doses of 3.6 mg once every 4 weeks or 10.8 mg once every 12 weeks. This study used US real-world evidence to characterize goserelin treatment patterns. METHODS: Electronic health record data of adults with a history of breast cancer and ≥2 goserelin prescriptions between January 1, 2017, and December 31, 2022, were identified through TriNetX. Patient demographics and treatment patterns were examined. RESULTS: Overall, 3,620 US patients were identified: 2,870 treated with goserelin 3.6 mg once every 4 weeks, 410 treated with 10.8 mg once every 12 weeks, and 340 switched from 3.6 mg once every 4 weeks to 10.8 mg once every 12 weeks. Peak utilization of 10.8 mg once every 12 weeks (36.6%) and dose switching to 10.8 mg once every 12 weeks (26.5%) occurred in 2020. Patients who switched to 10.8 mg once every 12 weeks had the longest median treatment duration (776 days), compared with the 3.6 mg once every 4 weeks and 10.8 mg once every 12 weeks cohorts (264 and 429 days, respectively). Of patients who switched, 65% were still being treated after 2 years, compared with 30% and 40% treated with 3.6 mg once every 4 weeks only or 10.8 mg once every 12 weeks only, respectively. Patients initially treated with or who switched to 10.8 mg once every 12 weeks were more adherent (64.4%-75.0%), compared with patients treated with 3.6 mg once every 4 weeks (45.4%). CONCLUSION: Treatment with goserelin 10.8 mg once every 12 weeks is associated with greater adherence and longer treatment duration, compared with 3.6 mg once every 4 weeks in patients with breast cancer in the United States.

  PublisherDOI

• Real-World Evidence of First Year Outcomes, Health Care Utilization, and Costs Among Kidney Transplant Recipients in the United States (USA), 2017-2023

  Journal of the American Society of Nephrology · 2025-10-01

  article
  PublisherDOI

• Treatment patterns of goserelin 3.6 mg once monthly and 10.8 mg every three months in women with breast cancer: A real-world analysis.

  JCO Oncology Practice · 2024-09-30

  article

  391 Background: Goserelin is a gonadotropin-releasing hormone agonist used for ovarian function suppression in the treatment of pre- and peri-menopausal patients with breast cancer. Goserelin, administered as a subcutaneous implant, is available in doses of 3.6 mg once monthly or 10.8 mg every three months; goserelin 3.6 mg is approved in the US, and both the 3.6 mg and 10.8 mg doses are approved by Health Canada and the European Medicines Agency. A study utilizing US real-world evidence was conducted to characterize treatment patterns of patients treated with goserelin 3.6 mg and/or 10.8 mg. Methods: Electronic health record data of adult patients with a history of breast cancer and with ≥2 prescriptions of goserelin between 1 January 2017 – 31 December 2022 were identified through TriNetX. Index date was the initiation of goserelin administration. Patients were followed until 15 March 2024. Patient demographics, treatment adherence, and healthcare resource utilization (HCRU) were examined and summarized using descriptive analytics. Results: Overall, 3,620 patients were identified: 2,870 treated with goserelin 3.6 mg, 410 with goserelin 10.8 mg, and 340 who switched from goserelin 3.6 mg to 10.8 mg. Peak utilization of goserelin 10.8 mg (36.6%) and peak dosage switching (26.5%) occurred in 2020. Across groups, mean age at index date was 42.2–44.4 years, and patients were white (64.1–67.6%), black (11.8%–13.9%), Asian (8.0%–11.8%), and American Indian and/or Alaskan native (0.3%–2.9%). Of patients with known BMI, BMI mostly ranged between 18.5–24.9 kg/m 2 in all groups. Patients who switched from goserelin 3.6 mg to 10.8 mg had the longest median treatment duration (776 days) (Table). Patients treated with goserelin 10.8 mg remained on treatment for a median of 426 days; 74.4% of these patients were treatment-adherent (Table). In comparison, patients treated with goserelin 3.6 mg had a shorter median treatment duration (226 days) and were less adherent (56.4%) (Table). HCRU during the 12 months after index date was broadly similar across groups. Conclusions: Treatment with goserelin 10.8 mg every three months is associated with greater adherence and longer treatment duration, compared with 3.6 mg once monthly, in patients with a history of breast cancer. Patients who switched from goserelin 3.6 mg to 10.8 mg were treatment-adherent for nearly two years, consistent with clinical recommendations. Goserelin 3.6 mg (n = 2,870) Goserelin 10.8 mg (n = 410) Switched from goserelin 3.6 mg to 10.8 mg (n = 340) Median duration of treatment (days) 264 429 776 Median time to switch (days) — — 218 Patients adherent* to dosing schedule 1,608 (56.4%) 306 (74.4%) — Status as of March 15, 2024 Discontinued goserelin 2,018 (70.8%) 261 (64.3%) 173 (50.9%) Death 88 (3.1%) 10 (2.5%) 3 (0.9%) *Defined as: ≤36 days between prescriptions (3.6 mg) or ≤108 days between prescriptions (10.8 mg).

  PublisherDOI

• Differentiable Uncalibrated Imaging

  IEEE Transactions on Computational Imaging · 2023-12-22 · 6 citations

  articleOpen access1st authorCorresponding

  We propose a differentiable imaging framework to address uncertainty in <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">measurement coordinates</i> such as sensor locations and projection angles. We formulate the problem as measurement interpolation at unknown nodes supervised through the forward operator. To solve it we apply implicit neural networks, also known as neural fields, which are naturally differentiable with respect to the input coordinates. We also develop differentiable spline interpolators which perform as well as neural networks, require less time to optimize and have well-understood properties. Differentiability is key as it allows us to jointly fit a measurement representation, optimize over the uncertain measurement coordinates, and perform image reconstruction which in turn ensures consistent calibration. We apply our approach to 2D and 3D computed tomography, and show that it produces improved reconstructions compared to baselines that do not account for the lack of calibration. The flexibility of the proposed framework makes it easy to extend to almost arbitrary imaging problems.

  PublisherOA PDFDOI

• 110 Deep learning models identify key tumor microenvironment features associated with genetic signatures of UV mutagenesis and alkylating agent treatment in melanoma

  Regular and Young Investigator Award Abstracts · 2023-10-31

  articleOpen access

  <h3>Background</h3> Melanoma is the most aggressive type of skin cancer and often exhibits therapeutic resistance.^{1 2} Different types of mutagenesis, for example UV exposure,^{3 4} have been shown to result in distinct genetic signatures; however, their impact on histological features of the tumor microenvironment (TME) and response to treatment remains unknown. Alkylating agents are one of the most commonly used chemo-therapeutics for melanoma⁵; however, the impact of alkylating agents on the melanoma TME is also poorly understood. In this work, we quantified the TME in melanoma using machine learning and investigated TME feature associations with 1) increased UV mutagenesis, and 2) alkylating agent-induced mutations. <h3>Methods</h3> PathExplore convolutional neural network-based models using hematoxylin and eosin (H&amp;E)-stained whole slide images (WSI) were trained to classify histologic substances in the TME (table 1). We quantified model performance using nested pairwise comparisons with pathologist annotation.⁶ We deployed PathExplore Melanoma along with a separately trained stromal subtyping model⁷ to extract human-interpretable features (HIFs) that quantify the TME across each WSI in the TCGA (SKCM, N=363) cohort. We identified mutational signatures indicative of UV and alkylating agents using the deconstructSigs R package.⁸ We utilized primary (N=71) and lymph node metastasis (N=255) slides for UV exposure analysis, and only primary slides for alkylating agent analysis. We quantified associations between HIFs and mutational signatures using univariate logistic regressions. P-values were corrected using Benjamini-Hochberg. Multivariable Cox models were used for survival analysis. <h3>Results</h3> We found a positive association between tumor-infiltrating lymphocyte (TIL) abundance (p=0.01), as well as the area proportion of densely inflamed stromal regions (p=0.015), with UV exposure. Features quantifying neutrophil abundance were associated with alkylating agent treatment, most notably neutrophil-to-lymphocyte ratio (NLR; p=0.013). Higher NLR was associated with worse overall survival in general, but this effect was attenuated in patients previously treated with alkylating agents. <h3>Conclusions</h3> We found that TIL abundance was associated with UV exposure, likely due to increased tumor mutational burden, which may have implications for immunotherapy. Additionally, NLR has previously been associated with poor prognosis in melanoma.^{9 10} Our results indicate that the effect of NLR on prognosis is also mediated by prior treatment, pointing to a complex causal web between TME, treatment, and patient outcomes. Broadly, these results suggest that machine learning can extract meaningful information regarding underlying mutation-driven or treatment-induced changes in the TME. <h3>References</h3> Kavran, Andrew J, <i>et al.</i> ‘Intermittent treatment of BRAFV600E melanoma cells delays resistance by adaptive resensitization to drug rechallenge.’ <i>Proceedings of the National Academy of Sciences</i> 2022;<b>119</b>(12):e2113535119. Rossi, Alessandro, <i>et al.</i> ‘Drug resistance of BRAF-mutant melanoma: Review of up-to-date mechanisms of action and promising targeted agents.’ <i>European journal of pharmacology</i> 2019;<b>862</b>:172621. Autier, Philippe, Jean-François Doré. ‘Ultraviolet radiation and cutaneous melanoma: a historical perspective.’ <i>Melanoma Research</i> 2020;<b>30</b>(2):113–125. Dousset, Léa, <i>et al.</i> ‘Positive association between location of melanoma, ultraviolet signature, tumor mutational burden, and response to anti-PD-1 therapy.’ <i>JCO precision oncology</i> 2021;<b>5</b>:1821–1829. Arozarena, Imanol, <i>et al.</i> ‘Differential chemosensitivity to antifolate drugs between RAS and BRAF melanoma cells.’ <i>Molecular Cancer</i> 2014;<b>13</b>(1):1–13. Gerardin, Ylaine, <i>et al.</i> ‘Improved statistical benchmarking of digital pathology models using pairwise frames evaluation.’ arXiv preprint arXiv:2306.04709 (2023). Najdawi, Fedaa, <i>et al.</i> ‘Artificial intelligence (AI)-based classification of stromal subtypes reveals associations between stromal composition and prognosis in NSCLC.’ <i>Cancer Research</i> 2023;<b>83</b>(7_Supplement):5447–5447. Rosenthal R, McGranahan N, Herrero J, Taylor BS, Swanton C. DeconstructSigs: delineating mutational processes in single tumors distinguishes DNA repair deficiencies and patterns of carcinoma evolution. <i>Genome Biol</i>. 2016 Feb 22;<b>17</b>:31. doi: 10.1186/s13059–016-0893–4. PMID: 26899170; PMCID: PMC4762164. Capone, Mariaelena, <i>et al.</i> ‘Baseline neutrophil-to-lymphocyte ratio (NLR) and derived NLR could predict overall survival in patients with advanced melanoma treated with nivolumab.’ <i>Journal for immunotherapy of cancer</i> 2018;<b>6</b>:1–7. Cohen, Joshua T, Thomas J Miner, Michael P Vezeridis. ‘Is the neutrophil-to-lymphocyte ratio a useful prognostic indicator in melanoma patients?.’ <i>Melanoma Management</i> 2020;<b>7</b>(3):MMT47.

  PublisherOA PDFDOI

• A Case of Lymphangioma of the Calf Region: Imaging Spectrum With Histopathological Correlation

  Cureus · 2023-11-14 · 3 citations

  articleOpen access1st authorCorresponding

  Lymphangioma, also known as cystic hygroma are benign malformations arising from abnormal development of the lymphatic system. Most often these lesions are found in the pediatric population, having a predilection for the neck/axilla, and are less common in extremities. Symptoms can vary based on size and location. Treatment is not usually indicated until they start impacting life due to deformity or symptoms such as pain, paraesthesia, etc. Here, we report a case report of lymphangioma located in the calf region of the right lower limb presenting in adult age.

  PublisherOA PDFDOI

Frequent coauthors

• Ivan Dokmanić

  19 shared

• Konik Kothari

  Google (United States)

  10 shared

• Bighnaraj Naik

  Veer Surendra Sai University of Technology

  9 shared

• Kumar Reddy

  9 shared

• Janmenjoy Nayak

  9 shared

• Danilo Pelusi

  University of Teramo

  9 shared

• Indrajeet Reza

  University of Teramo

  9 shared

• Motahar Padhi

  Veer Surendra Sai University of Technology

  9 shared

Labs

• Siebel School of Computing and Data SciencePI

Education

• Doctor of Philosophy, Electrical and Computer Engineering

  University of Illinois at Urbana-Champaign

• Bachelor of Arts, Engineering

  University of Cambridge

  2014

• Master of Engineering, Engineering

  University of Cambridge

  2014

Awards & honors

• NSF CAREER award to improve robot navigation and object inte…