Saimourya Surabhi

Mrs. Saimourya Surabhi

orcid

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

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

Representative Publications

View All Publications
21 March 2023 in JMIR Formative Research
Verified
Journal Article
Open access
ABAgnik Banerjee
Agnik Banerjee
OMOnur Cezmi Mutlu
Onur Cezmi Mutlu
AKAaron Kline
Aaron Kline
SSSaimourya Surabhi
Saimourya Surabhi
Peter WashingtonPeter Washington
Peter Washington
DWDennis Wall
Dennis Wall
Background: Implementing automated facial expression recognition on mobile devices could provide an accessible diagnostic and therapeutic tool for those who struggle to recognize facial expressions, including children with developmental behavioral conditions such as autism. Despite recent advances in facial expression classifiers for children, existing models are too computationally expensive for smartphone use. Objective: We explored several state-of-the-art facial expression classifiers designed for mobile devices, used posttraining optimization techniques for both classification performance and efficiency on a Motorola Moto G6 phone, evaluated the importance of training our classifiers on children versus adults, and evaluated the models’ performance against different ethnic groups. Methods: We collected images from 12 public data sets and used video frames crowdsourced from the GuessWhat app to train our classifiers. All images were annotated for 7 expressions: neutral, fear, happiness, sadness, surprise, anger, and disgust. We tested 3 copies for each of 5 different convolutional neural network architectures: MobileNetV3-Small 1.0x, MobileNetV2 1.0x, EfficientNetB0, MobileNetV3-Large 1.0x, and NASNetMobile. We trained the first copy on images of children, second copy on images of adults, and third copy on all data sets. We evaluated each model against the entire Child Affective Facial Expression (CAFE) set and by ethnicity. We performed weight pruning, weight clustering, and quantize-aware training when possible and profiled each model’s performance on the Moto G6. Results: Our best model, a MobileNetV3-Large network pretrained on ImageNet, achieved 65.78% accuracy and 65.31% F1-score on the CAFE and a 90-millisecond inference latency on a Moto G6 phone when trained on all data. This accuracy is only 1.12% lower than the current state of the art for CAFE, a model with 13.91x more parameters that was unable to run on the Moto G6 due to its size, even when fully optimized. When trained solely on children, this model achieved 60.57% accuracy and 60.29% F1-score. When trained only on adults, the model received 53.36% accuracy and 53.10% F1-score. Although the MobileNetV3-Large trained on all data sets achieved nearly a 60% F1-score across all ethnicities, the data sets for South Asian and African American children achieved lower accuracy (as much as 11.56%) and F1-score (as much as 11.25%) than other groups. Conclusions: With specialized design and optimization techniques, facial expression classifiers can become lightweight enough to run on mobile devices and achieve state-of-the-art performance. There is potentially a “data shift” phenomenon between facial expressions of children compared with adults; our classifiers performed much better when trained on children. Certain underrepresented ethnic groups (e.g., South Asian and African American) also perform significantly worse than groups such as European Caucasian despite similar data quality. Our models can be integrated into mobile health therapies to help diagnose autism spectrum disorder and provide targeted therapeutic treatment to children.

21 October 2025 in JMIR AI
Verified
Journal Article
Open access
KLKaiying Kevin Lin
Kaiying Kevin Lin
Abdur RasoolAbdur Rasool
Abdur Rasool
SSSaimourya Surabhi
Saimourya Surabhi
CMCezmi Mutlu
Cezmi Mutlu
HZHaopeng Zhang
Haopeng Zhang
Peter WashingtonPeter Washington
Peter Washington
Background: Large language models (LLMs) have demonstrated the ability to perform complex tasks traditionally requiring human intelligence. However, their use in automated diagnostics for psychiatry and behavioral sciences remains under-studied. Objective: This study aimed to evaluate whether incorporating structured clinical assessment scales improved the diagnostic performance of LLM-based chatbots for neuropsychiatric conditions (we evaluated autism spectrum disorder, aphasia, and depression datasets) across two prompting strategies: (1) direct diagnosis and (2) code generation. We aimed to contextualize LLM-based diagnostic performance by benchmarking it against prior work that applied traditional machine learning classifiers to the same datasets, allowing us to assess whether LLMs offer competitive or complementary capabilities in clinical classification tasks. Methods: We tested two approaches using ChatGPT, Gemini, and Claude models: (1) direct diagnostic querying and (2) execution of chatbot-generated code for classification. Three diagnostic datasets were used: ASDBank (autism spectrum disorder), AphasiaBank (aphasia), and Distress Analysis Interview Corpus-Wizard-of-Oz interviews (depression and related conditions). Each approach was evaluated with and without the aid of clinical assessment scales. Performance was compared to existing machine learning benchmarks on these datasets. Results: Across all 3 datasets, incorporating clinical assessment scales led to little improvement in performance, and results remained inconsistent and generally below those reported in previous studies. On the AphasiaBank dataset, the direct diagnosis approach using ChatGPT with GPT-4 produced a low F1-score of 65.6% and specificity of 33%. The code generation method improved results, with ChatGPT with GPT-4o reaching an F1-score of 81.4%, specificity of 78.6%, and sensitivity of 84.3%. ChatGPT with GPT-o3 and Gemini 2.5 Pro performed even better, with F1-scores of 86.5% and 84.3%, respectively. For the ASDBank dataset, direct diagnosis results were lower, with F1-scores of 56% for ChatGPT with GPT-4 and 54% for ChatGPT with GPT-4o. Under code generation, ChatGPT with GPT-o3 reached 67.9%, and Claude 3.5 performed reasonably well with 60%. Gemini 2.5 Pro failed to respond under this assessment condition. In the Distress Analysis Interview Corpus-Wizard-of-Oz dataset, direct diagnosis yielded high accuracy (70.9%) but poor F1-scores of 8% using ChatGPT with GPT-4o. Code generation improved specificity—88.6% with ChatGPT with GPT-4o—but F1-scores remained low overall. These findings suggest that, while clinical scales may help structure outputs, prompting alone remains insufficient for consistent diagnostic accuracy. Conclusions: Current LLM-based chatbots, when prompted naively, underperform on psychiatric and behavioral diagnostic tasks compared to specialized machine learning models. Clinical assessment scales might modestly aid chatbot performance, but more sophisticated prompt engineering and domain integration are likely required to reach clinically actionable standards.

30 May 2022 in JMIR Biomedical Engineering
Verified
Journal Article
Open access
ESEssam Sleiman
Essam Sleiman
OMOnur Cezmi Mutlu
Onur Cezmi Mutlu
SSSaimourya Surabhi
Saimourya Surabhi
AHArman Husic
Arman Husic
AKAaron Kline
Aaron Kline
DWDennis Wall
Dennis Wall
Journal of Medical Internet Research - International Scientific Journal for Medical Research, Information and Communication on the Internet #Preprint #PeerReviewMe: Warning: This is a unreviewed preprint. Readers are warned that the document has not been peer-reviewed by expert/patient reviewers or an academic editor, may contain misleading claims, and is likely to undergo changes before final publication, if accepted, or may have been rejected/withdrawn. Readers with interest and expertise are encouraged to sign up as peer-reviewer, if the paper is within an open peer-review period. Please cite this preprint only for review purposes or for grant applications and CVs (if you are the author). Background: Autism spectrum disorder (ASD) is a prevalent neurodevelopmental disorder encountered by 1 in 44 children in the United States. Patients with autism face difficulty effectively communicating with peers, articulating feelings and emotions, and controlling behaviors. Rapid and early diagnosis leads to improved treatment outcomes; however, current medical techniques often take years. Difficult, time-consuming diagnoses accompanied by the recent advances in computer vision have led to a surge of interest from researchers in developing models to streamline the autism diagnosis process. Objective: We aimed to assess the viability of at-home autism diagnosis by leveraging video data collected from a mobile game app to train a computer vision model using insights extracted from the change in emotion-related expression features over time. Methods: With our GuessWhat game-based mobile app, we collected a video data set of 74 children with ASD and neurotypical diagnoses actively playing in a natural home environment. To investigate and quantify the significance of facial emotion features for ASD detection, we developed a deep learning–based autism classifier with the following two components: a convolutional neural network (CNN) attached to a long short-term memory (LSTM). We pretrained our CNN backbone in the following two ways: (1) on ImageNet or (2) on a compilation of facial expression recognition data sets to analyze autism detection performance improvements using emotion features. The output of each CNN is fed into an LSTM model to diagnose ASD from video data. Results: Our top-performing architecture employing a CNN backbone trained on the facial expression recognition data sets and a unidirectional LSTM achieved a top accuracy of 91.2%, precision of 91.2%, recall of 89.8%, and an F1-score of 90.6%. Conclusions: We discovered the change in an individual’s facial expression features over time as a relevant marker for autism detection. Extracting emotional features from each frame resulted in a 27.8% F1-score improvement when compared to ImageNet weights. Our study demonstrates the capability of mobile apps to collect a natural, diverse data set for an improved autism diagnosis. These results demonstrate that deep learning and computer vision–based methods are instrumental for automated autism diagnosis from at-home recorded videos using unspecialized equipment.

22 November 2022 in arXiv
Verified
Preprint
Open access
PDF file
ALAnish Lakkapragada
Anish Lakkapragada
ESEssam Sleiman
Essam Sleiman
SSSaimourya Surabhi
Saimourya Surabhi
DWDennis Wall
Dennis Wall
Multi-Task Learning (MTL) is a growing subject of interest in deep learning, due to its ability to train models more efficiently on multiple tasks compared to using a group of conventional single-task models. However, MTL can be impractical as certain tasks can dominate training and hurt performance in others, thus making some tasks perform better in a single-task model compared to a multi-task one. Such problems are broadly classified as negative transfer, and many prior approaches in the literature have been made to mitigate these issues. One such current approach to alleviate negative transfer is to weight each of the losses so that they are on the same scale. Whereas current loss balancing approaches rely on either optimization or complex numerical analysis, none directly scale the losses based on their observed magnitudes. We propose multiple techniques for loss balancing based on scaling by the exponential moving average and benchmark them against current best-performing methods on three established datasets. On these datasets, they achieve comparable, if not higher, performance compared to current best-performing methods.

2 December 2025 in Algorithms
Verified
Journal Article
Open access
PDF file
MHMarie Amale Huynh
Marie Amale Huynh
AKAaron Kline
Aaron Kline
SSSaimourya Surabhi
Saimourya Surabhi
KDKaitlyn Dunlap
Kaitlyn Dunlap
OMOnur Cezmi Mutlu
Onur Cezmi Mutlu
DWDennis Wall
Dennis Wall
Early detection of Autism Spectrum Disorder (ASD), a neurodevelopmental condition characterized by social communication challenges, is essential for timely intervention. Naturalistic home videos collected via mobile applications offer scalable opportunities for digital diagnostics. We leveraged GuessWhat, a mobile game designed to engage parents and children, which has generated over 3000 structured videos from 382 children. From this collection, we curated a final analytic sample of 688 feature-rich videos centered on a single dyad, enabling more consistent modeling. We developed a two-step pipeline: (1) filtering to isolate high-quality videos, and (2) feature engineering to extract interpretable behavioral signals. Unimodal LSTM-based models trained on eye gaze, head position, and facial expression achieved test AUCs of 86% (95% CI: 0.79–0.92), 78% (95% CI: 0.69–0.86), and 67% (95% CI: 0.55–0.78), respectively. Late-stage fusion of unimodal outputs significantly improved predictive performance, yielding a test AUC of 90% (95% CI: 0.84–0.95). Our findings demonstrate the complementary value of distinct behavioral channels and support the feasibility of using mobile-captured videos for detecting clinically relevant signals. While further work is needed to improve generalizability and inclusivity, this study highlights the promise of real-time, scalable autism phenotyping for early interventions.

14 December 2025 in Biocomputing 2020
Verified
Conference Paper
Open access
AKAiden Ko
Aiden Ko
AKAaron Kline
Aaron Kline
KDKaitlyn Dunlap
Kaitlyn Dunlap
SSSaimourya Surabhi
Saimourya Surabhi
PAParnian Azizian
Parnian Azizian
DWDennis Wall
Dennis Wall
Uncertainty quantification remains an underdeveloped aspect of AI-based clinical decision tools. As AI systems become increasingly prevalent in healthcare, it is essential not only to measure uncertainty but also to manage it in ways that support clinical decision-making. In this study, we investigate abstention as a practical mechanism for managing uncertainty in diagnostic classifiers. To stress-test this approach, we deliberately evaluate abstention performance on a purposefully noisy dataset of pediatric autism video assessments comprising heterogeneous video sources and a diverse range of human raters. We apply abstention strategies to existing autism classifiers trained on diagnostic assessment data, comparing baseline performance to a range of thresholding configurations that trade off retained sample coverage against key clinical metrics. We compare performance gains from prioritizing sensitivity or specificity to targeting a balanced increase in Youden's J to demonstrate a wide variety of use cases that abstention can enable. This work demonstrates a concrete use case of introducing abstention into the output range of clinical decision models, enabling both uncertainty quantification and management in diagnostic classifiers.