About

Vijaya B. Kolachalama is an affiliate faculty member and assistant professor at Boston University School of Medicine. His laboratory focuses on developing advanced machine learning algorithms to address complex biomedical datasets, with a particular emphasis on medical imaging. His research includes phenotyping neurodegeneration through machine learning frameworks that process multimodal data and identify specific signatures of neurodegeneration, utilizing large data cohorts such as the Framingham Heart Study. Additionally, he develops computational frameworks based on deep learning to assist pathologists in digital pathology applications across kidney disease, lung cancer, and colorectal cancer. His work also involves creating machine learning frameworks to analyze large-scale studies related to musculoskeletal diseases, such as osteoarthritis, aiming to quantify structures responsible for pain and factors contributing to disease progression.

Research topics

• Artificial Intelligence
• Computer Science
• Pathology
• Psychiatry
• Natural Language Processing
• Psychology
• Medicine
• Clinical psychology
• Biology
• Anatomy
• Neuroscience
• Theoretical computer science

Selected publications

• Additional file 1 of Attention-based deep learning for analysis of pathology images and gene expression data in lung squamous premalignant lesions

  Figshare · 2026-04-08

  articleOpen access

  Additional file 1: Figure S1. Whole slide image pre-processing pipeline. Figure S2. Gene expression data pre-processing pipeline. Figure S3. Data split and cross-validation scheme. Figure S4. Heatmap of statistical significance between prediction probabilities grouped by histology grade. Figure S5. Scatter plot of external testing sample prediction probabilities. Figure S6. Principal Component (PC) Analysis plots on external testing samples by cohort (extended from Fig. 2H and Fig. 3G). Figure S7. Gene heatmap of external testing biopsy samples. Table S1. Comparison of model performance with external multimodal models. Table S2. Accuracy, sensitivity, and specificity across histologic grades.

  PublisherDOI

• Domain-adapted language model using reinforcement learning for various dementias

  medRxiv · 2026-03-23

  articleOpen accessSenior authorCorresponding

  Large language models excel at processing complex clinical data and advanced reasoning, yet domain-specific adaptation is essential to realize their full potential in fields such as Alzheimer's disease and related dementias (ADRD). Here, we present a generative language model for ADRD fine-tuned via reinforcement learning with verifiable rewards using a self-certainty-aware advantage. Model development and validation leveraged data from five ADRD cohorts, totaling 54, 535 participants. Our framework integrates demographics, personal and family medical histories, medication use, neuropsychological test results, functional assessments, physical and neurological examination findings, laboratory data and multimodal neuroimaging to construct comprehensive clinical profiles. On held-out testing data involving 36, 688 participants, our model achieved robust performance on syndromic classification, primary etiological diagnosis and biomarker prediction. Model predictions were validated against postmortem-confirmed diagnoses, and clinical utility was demonstrated in a controlled within-subjects crossover study where board-certified neurologists reviewed cases with and without model assistance, showing that exposure to model responses improved diagnostic performance. These results demonstrate that targeted domain adaptation with reinforcement learning can enable language models to deliver accurate, reasoning-driven support in ADRD evaluation. Prospective validation will be essential to translate these advances into improved patient outcomes.

  PublisherOA PDFDOI

• Intravascular Ultrasound is More Accurate Than Angiography in Arteriovenous Vascular Access Lesions

  Kidney360 · 2026-02-13

  articleOpen access

  BACKGROUND: Conventional angiography remains the standard diagnostic modality for arteriovenous (AV) access dysfunction in hemodialysis patients, but its geometric accuracy is limited. Intravascular ultrasound (IVUS) offers superior lesion detection, yet its absolute measurement accuracy remains uncertain. Using three-dimensional (3D) printed vascular conduits as reference standards, we assessed the accuracy of IVUS versus angiography, hypothesizing that complex conduit geometry, quantified by Gaussian curvature, would exacerbate angiographic error. METHODS: Clinically relevant AV access geometries were modeled with computer-aided design (CAD) and fabricated using 3D printing. Lumen diameters were measured by contrast angiography and IVUS and compared with CAD dimensions. Conduit geometry was characterized using finite element-based Gaussian curvature mapping. Paired Student's t-test, Tukey-Kramer correction for multiple testing, and linear mixed-effect modeling were performed to examine the influence of clustering of lesions within conduits. RESULTS: IVUS demonstrated significantly lower measurement error compared with angiography, especially in stenotic segments with >50% luminal narrowing, which persisted even after correction for multiple comparisons (P < 0.05). These high-grade stenoses frequently coincided with regions of especially high positive or negative Gaussian curvature, reflecting complex conduit geometry. In such regions, angiography consistently underestimated lumen diameter, with error magnitude increasing in curved or tortuous lesions. IVUS measurements closely approximated CAD measurements, retaining accuracy even in severe stenoses. For mild stenoses (<50%) and aneurysmal dilatations, both modalities performed comparably. CONCLUSIONS: Geometric complexity directly contributes to modality-specific error. Angiography systematically underestimates lumen dimensions in complex, stenotic regions, while IVUS detects them with higher fidelity and preserves accuracy. These findings establish IVUS as the more reliable modality for evaluating AV access dysfunction and support its integration into routine practice in guiding intervention for AV access stenosis.

  PublisherDOI

• Tryptophan metabolism reprogramming potentially contributes to the prothrombotic milieu in mice and humans with SARS-CoV-2

  Blood Vessels Thrombosis & Hemostasis · 2026-01-05

  articleOpen access

  Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection disturbs the coagulation balance in the blood, triggering thrombosis and contributing to organ failure. The role of prothrombotic metabolites in COVID-19-associated coagulopathy remains elusive. Leveraging K18-hACE2 mice infected with SARS-CoV-2, we observed higher levels of the tryptophan metabolite, kynurenine, than in controls. SARS-CoV-2-infected mice showed a significant upregulation of enzymes controlling kynurenine biogenesis, such as indoleamine 2,3-dioxygenase 1 (IDO-1) and tryptophan 2,3-dioxygenase in the kidney and liver, respectively, as well as changes in the enzymes involved in kynurenine catabolism, including kynurenine monooxygenase and kynurinase. Consistent with the agonistic role of these metabolites in aryl hydrocarbon receptor (AHR) signaling, AHR activation and its downstream mediator, tissue factor (TF), a highly potent procoagulant factor, was observed in endothelial cells (ECs) of lungs and kidneys of infected mice. These findings were validated in humans. Compared with controls, sera of patients with COVID-19 showed increased levels of kynurenine, kynurenic acid, anthranilic acid, and quinolinic acid. Activation of the AHR-TF axis was noted in the kidneys and lungs of patients with COVID-19, and sera from patients infected with SARS-CoV-2 showed higher IDO-1 activity than controls. Kynurenine levels in patients with COVID-19 correlated strongly with the TF-inducing activity of sera from patients infected with SARS-CoV-2 on ECs. A specific IDO-1 inhibitor or AHR inhibitor separately or in combination suppressed sera from induced TF activity in ECs from patients with COVID-19. Together, we identified IDO-1 as upregulated by SARS-CoV-2 infection, resulting in augmented kynurenine and its prothrombotic catabolites, thereby suggesting the kynurenine-AHR-TF axis as a potential new diagnostic and therapeutic target.

  PublisherDOI

• Additional file 2 of Attention-based deep learning for analysis of pathology images and gene expression data in lung squamous premalignant lesions

  Figshare · 2026-04-08

  datasetOpen access

  Additional file 2: Complete lists of selected up- and down-regulated genes in each fold.

  PublisherDOI

• Additional file 1 of Attention-based deep learning for analysis of pathology images and gene expression data in lung squamous premalignant lesions

  Figshare · 2026-04-08

  articleOpen access

  Additional file 1: Figure S1. Whole slide image pre-processing pipeline. Figure S2. Gene expression data pre-processing pipeline. Figure S3. Data split and cross-validation scheme. Figure S4. Heatmap of statistical significance between prediction probabilities grouped by histology grade. Figure S5. Scatter plot of external testing sample prediction probabilities. Figure S6. Principal Component (PC) Analysis plots on external testing samples by cohort (extended from Fig. 2H and Fig. 3G). Figure S7. Gene heatmap of external testing biopsy samples. Table S1. Comparison of model performance with external multimodal models. Table S2. Accuracy, sensitivity, and specificity across histologic grades.

  PublisherDOI

• Additional file 2 of Attention-based deep learning for analysis of pathology images and gene expression data in lung squamous premalignant lesions

  Figshare · 2026-04-08

  datasetOpen access

  Additional file 2: Complete lists of selected up- and down-regulated genes in each fold.

  PublisherDOI

• Performance of open-source large language models on nephrology self-assessment program

  medRxiv · 2026-04-16

  articleOpen accessSenior author

  Background: Large Language Models (LLMs) have demonstrated strong performance in medical question-answering tasks, highlighting their potential for clinical decision support and medical education. However, their effectiveness in subspecialty areas such as nephrology remains underexplored. In this study, we assess the performance of open-source LLMs in answering multiple-choice questions from the Nephrology Self-Assessment Program (NephSAP) to better understand their capabilities and limitations within this specialized clinical domain. Methods: We evaluated the performance of five open-source large language models (LLMs): PodGPT which a podcast-pretrained model focused on STEMM disciplines, Llama 3.2-11B, Mistral-7B-Instruct-v0.2, Falcon3-10B-Instruct, and Gemma-2-9B-it. Each model was tested on its ability to answer multiple-choice questions derived from the NephSAP. Model performance was quantified using accuracy, defined as the proportion of correctly answered questions. In addition, the quality of the models' explanatory responses was assessed using several natural language processing (NLP) metrics: Bilingual Evaluation Understudy (BLEU), Word Error Rate (WER), cosine similarity, and Flesch-Kincaid Grade Level (FKGL). For qualitative analysis, three board-certified nephrologists reviewed 40 randomly selected model responses to identify factual and clinical reasoning errors, with performance summarized as average error ratios based on the proportion of error-associated words per response. Results: Among the evaluated models, PodGPT achieved the highest accuracy (64.77%), whereas Llama showed the lowest performance with an accuracy of 45.08%. Qualitative analysis showed that PodGPT had the lowest factual error rate (0.017), while Llama and Falcon achieved the lowest reasoning error rates (0.038). Conclusions: This study highlights the importance of STEMM-based training to enhance the reasoning capabilities and reliability of LLMs in clinical contexts, supporting the development of more effective AI-driven decision-support tools in nephrology and other medical specialties.

  PublisherDOI

• Higher-Order Domain Generalization in Magnetic Resonance-Based Assessment of Alzheimer's Disease

  arXiv (Cornell University) · 2026-01-04

  preprintOpen access

  Despite progress in deep learning for Alzheimer's disease (AD) diagnostics, models trained on structural magnetic resonance imaging (sMRI) often do not perform well when applied to new cohorts due to domain shifts from varying scanners, protocols and patient demographics. AD, the primary driver of dementia, manifests through progressive cognitive and neuroanatomical changes like atrophy and ventricular expansion, making robust, generalizable classification essential for real-world use. While convolutional neural networks and transformers have advanced feature extraction via attention and fusion techniques, single-domain generalization (SDG) remains underexplored yet critical, given the fragmented nature of AD datasets. To bridge this gap, we introduce Extended MixStyle (EM), a framework for blending higher-order feature moments (skewness and kurtosis) to mimic diverse distributional variations. Trained on sMRI data from the National Alzheimer's Coordinating Center (NACC; n=4,647) to differentiate persons with normal cognition (NC) from those with mild cognitive impairment (MCI) or AD and tested on three unseen cohorts (total n=3,126), EM yields enhanced cross-domain performance, improving macro-F1 on average by 2.4 percentage points over state-of-the-art SDG benchmarks, underscoring its promise for invariant, reliable AD detection in heterogeneous real-world settings. The source code will be made available upon acceptance at https://github.com/zobia111/Extended-Mixstyle.

  PublisherDOI

• Vision-language framework for multi-sequence brain magnetic resonance imaging

  medRxiv · 2026-04-04

  articleOpen accessSenior authorCorresponding

  Structural magnetic resonance imaging (MRI) is a cornerstone for diagnosing neurological disorders, yet automated interpretation of multi-sequence brain MRI remains limited by challenges in cross-sequence reasoning and protocol variability. Here we present ReMIND, a vision-language modeling framework tailored for comprehensive multi-sequence and multi-volumetric brain MRI analysis. Trained on over 73,000 deidentified patient visits encompassing more than 850,000 MRI sequences paired with radiology reports from diverse clinical and research cohorts, ReMIND combined large-scale instruction tuning on more than one million clinically grounded question-answer (QA) pairs with targeted supervised fine-tuning for radiology report generation. At inference, ReMIND employed modality-aware reranking and correction, a report-level decoding strategy that suppressed unsupported modality claims while preserving linguistic fluency and clinical coherence. Cross-cohort generalization was maintained on independent external datasets from different institutions. These findings represent an advance toward consistent and equitable brain MRI interpretation, meriting prospective evaluation to support diagnosis and management of neurological conditions.

  PublisherOA PDFDOI

Recent grants

• PRISTINE: Pre-cancer histology identification of Endobronchial biopsies using deep learning

  NIH · $424k · 2020–2024

• Smartphone image analysis for real time adequacy assessment during kidney biopsy

  NIH · $252k · 2022–2024

• Mechanisms of drug-coated balloon therapy

  NIH · $2.1M · 2021–2026

• Cognitive Heterogeneity in those with high Alzheimer's Disease Risk

  NIH · $1.8M · 2020–2023

Frequent coauthors

• Rhoda Au

  Boston University

  158 shared

• Elazer R. Edelman

  151 shared

• Ileana De Anda‐Duran

  67 shared

• Kumaran Kolandaivelu

  Brigham and Women's Hospital

  65 shared

• Lydia Bazzano

  57 shared

• David J. Libon

  Rowan University

  54 shared

• Owen Carmichael

  Pennington Biomedical Research Center

  54 shared

• Shakira F. Suglia

  Emory University

  50 shared

Education

• Ph.D., Physics

  University of California, Berkeley

  1996

• M.S., Physics

  University of California, Berkeley

  1993

• B.S., Physics

  University of Madras

  1991