About

Iain Carmichael is an Assistant Professor of Pathology and Data Science at UNC-Chapel Hill and a Visiting Assistant Professor in the Department of Pathology at UCSF. He leads an interdisciplinary computational pathology group focused on building data-driven, computational systems to analyze high-resolution histology images of diseased tissue alongside other clinical data sources. His research aims to leverage artificial intelligence to improve diagnostic precision, reduce healthcare costs, accelerate clinical workflows, and expand access to diagnostic expertise, particularly in the context of cancer and other diseases. The group’s work also seeks to facilitate the discovery of novel biomarkers for disease prognosis and therapeutic response, as well as to advance basic scientific understanding of disease processes through computational approaches. The major focus of Carmichael's research is the analysis of large-scale, high-resolution spatial-omics data generated in pathology laboratories, including digitized H&E stained slides, multiplex immunofluorescence, spatially resolved transcriptomics, and 3D histology images. These data sets are characterized by their immense scale and complexity, often requiring the development of new algorithms to fully exploit emerging tissue measurement technologies. His work addresses the challenges of integrating diverse tissue-based data modalities with molecular, radiology, and electronic health record information, which presents significant opportunities for biomedical discovery alongside substantial statistical and computational challenges. To tackle these problems, Carmichael’s group combines domain expertise with deep learning, computer vision, statistical inference, and open-source software development.

Research topics

• Artificial Intelligence
• Machine Learning
• Computer Science
• Clinical psychology
• Medicine
• Genetics
• Demography
• Biology
• Internal medicine
• Psychiatry

Selected publications

• 533 AI-Based Bone Marrow Aspirate Differentials with Real-World Distribution Shift Calibration

  Laboratory Investigation · 2026-03-01

  article
  PublisherDOI

• Perceived Importance of Counseling Among Patients Receiving Methadone Treatment

  Journal of Psychoactive Drugs · 2025-03-27

  article

  < .001. We conclude that most participants perceived counseling to be important and OUD treatment beliefs independently predicted perceived importance of counseling.

  PublisherDOI

• DeepHeme, a high-performance, generalizable deep ensemble for bone marrow morphometry and hematologic diagnosis

  Science Translational Medicine · 2025-06-11 · 4 citations

  article

  Cytomorphological analysis of the bone marrow aspirate (BMA) is pivotal for the diagnostic workup of a broad range of hematological disorders. However, this skill is error prone, highly complex, and time consuming. Deep learning-based models for the automatic classification of bone marrow cell morphology demonstrate the potential to improve diagnostic efficiency and accuracy. However, existing deep learning approaches in this field fall short of expert-level performance and lack generalizability beyond a single dataset. Working with multiple hematopathologists, we curated a dataset from the University of California, San Francisco, which included a training set of 30,394 images from 40 patients with morphologically normal marrows and a test set of 8507 images from 10 different patients, all derived from 400×-equivalent whole-slide images (WSIs). We then developed DeepHeme, a snapshot ensemble deep learning classifier, which outperformed previous models in accuracy while expanding the total number of differentiable cell classes. We externally validated DeepHeme using an independent dataset from the Memorial Sloan Kettering Cancer Center, which included 2694 images from 10 morphologically normal patients and 11,076 images from 655 patients with normal or diseased marrow, scanned using a different WSI system, demonstrating robust generalizability. At the level of individual cell classifications, we systematically compared DeepHeme's diagnostic performance with that of three medical experts from different academic hospitals, demonstrating that DeepHeme achieved accuracy comparable to, or exceeding, that of human experts. Accurate and generalizable cell classification represents a step toward automated analysis of hematopathology slides and the development of quantitative, morphology-based, predictive markers.

  PublisherDOI

• Measure of Strength of Evidence for Visually Observed Differences between Subpopulations

  UNC Libraries · 2025-07-25

  articleOpen accessSenior author

  For measuring the strength of visually observed subpopulation differences, the Population Difference Criterion is proposed to assess the statistical significance of visually observed subpopulation differences. It addresses the following challenges: in high-dimensional contexts, distributional models can be dubious; in high-signal contexts, conventional permutation tests give poor pairwise comparisons. We also make two other contributions: Based on a careful analysis we find that a balanced permutation approach is more powerful in high-signal contexts than conventional permutations. Another contribution is the quantification of uncertainty due to permutation variation via a bootstrap confidence interval. The practical usefulness of these ideas is illustrated in the comparison of subpopulations of modern cancer data.

  PublisherDOI

• Interpretable Multiple Instance Learning for Hematologic Diagnosis from Peripheral Blood Smears

  Communications Medicine · 2025-10-31

  preprintOpen access

  Accurate diagnosis of hematologic malignancies from peripheral blood smears (PBSs) requires integrating cellular morphology and composition across hundreds of white blood cells. Existing approaches primarily automate single-cell classification and do not provide whole-slide diagnostic predictions. We present a full network that utilizes a highly performative cell-based encoder (DeepHeme) for feature extraction paired with our weakly supervised framework using attention-based multiple instance learning (MIL) that we call CAREMIL (Cell AggRegation, Explainable, Multiple Instance Learning). Upon evaluating various popular image encoders and MIL architectures, the combination of DeepHeme and CAREMIL is the best performing pipeline on our disease classification task. CAREMIL proves to be a robust aggregation function that outperforms the most commonly used slide level aggregation function (gated multiple instance learning) across several encoder types. The greatest improvements in performance gain with CAREMIL is observed when using out-of-domain encoders, including an encoder trained on ImageNet and leading open-source pathology foundational models (UNI2 and Virchow2). CAREMIL plus DeepHeme achieves the highest diagnostic performance across acute leukemia (AML), myelodysplastic syndromes (MDS), and hairy cell leukemia (HCL) (AUROCs 0.999, 0.891, and 0.945, respectively), and identifies AML disease even in cases with minimal or absent circulating blasts. Attention values assigned by CAREMIL highlight diagnostically relevant cells and reveal disease-specific morphometric signatures, enabling biological interpretability and case-level insight. CAREMIL remains robust to misclassified cell types by the cell image encoder and does not require explicit cell-level supervision. These findings position CAREMIL as an effective and interpretable multiple instance learning framework for hematologic slide diagnosis, with potential to extend to bone marrow aspirates, cytology, and other liquid biopsy specimens, and to support a broader shift toward quantitative, morphology-informed diagnostics in hematology.

  PublisherDOI

• A Quantitative Examination of Illness Models Among People With Opioid Use Disorder Receiving Methadone Treatment

  Journal of Addiction Medicine · 2024-03-06 · 1 citations

  articleOpen access

  BACKGROUND: Few studies have examined illness models among people with addiction. We investigated illness models and their associations with demographics and treatment beliefs among patients receiving methadone treatment for opioid use disorder. METHODS: From January 2019 to February 2020, patients receiving methadone treatment at outpatient opioid treatment programs provided demographics and rated using 1 to 7 Likert-type scales agreement with addiction illness models (brain disease model, chronic medical condition model [CMCM], and no explanation [NEM]) and treatment beliefs. Pairwise comparisons and multivariate regressions were used to examine associations between illness models, demographics, and treatment beliefs. Statistical significance was set at P < 0.05. RESULTS: A total of 450 patients participated in the study. Forty percent self-identified as female, 13% as Hispanic, and 78% as White; mean age was 38.5 years. Brain disease model was the most frequently endorsed illness model (46.2%), followed by CMCM (41.7%) and NEM (21.9%). In multivariate analyses, agreement with brain disease model was significantly positively associated with beliefs that methadone treatment would be effective, counseling is important, and methadone is lifesaving, whereas agreement with CMCM was significantly positively associated with beliefs that methadone treatment would be effective, counseling is important, 12-step is the best treatment, taking methadone daily is important, and methadone is lifesaving. In multivariate analyses, agreement with NEM was negatively significantly associated with beliefs that methadone would be effective, counseling is important, taking methadone daily is important, and methadone is lifesaving. DISCUSSION: Many patients in methadone treatment endorsed medicalized addiction models. Agreement with addiction illness models appear to be related to treatment beliefs.

  PublisherOA PDFDOI

• Data science vs. statistics: two cultures?

  UNC Libraries · 2024-08-14 · 1 citations

  articleOpen accessSenior author
  PublisherDOI

• Deep Learning for Morphology-Based, Bone Marrow Cell Classification

  Blood · 2023-11-02 · 1 citations

  article

  The morphological classification of cells in bone marrow aspirate (BMA) is central to the diagnosis of hematologic diseases, including leukemias. Despite being a critical task, its monotonous, time-consuming nature and dependency on highly skilled clinical experts makes it prone to human error. Such errors can lead to delays and misdiagnoses that negatively impact patient care. To counter these challenges, we curated an expansive dataset of more than 40,000 hematopathologist consensus-annotated single-cell images, extracted from BMA whole slide images (WSIs), each annotated into one of 23 distinct morphologic classes. We then utilized this data to develop DeepHeme, a convolutional neural network classifier designed for bone marrow cell typing tasks. DeepHeme achieves state-of-the-art performance in both the breadth of differentiable classes and accuracy across these classes. By comparing its performance to that of individual hematopathologists from three premier academic medical centers, using our gold standard consensus-labelled images, we found our AI algorithm either matched or surpassed the average performance across all classes. In addition, we integrated DeepHeme with internally developed region classifier and cell detection algorithms, culminating in a comprehensive diagnostic pipeline for whole slide cell differential. We next tested DeepHeme on slides from an external hospital system at a major cancer center to evaluate the generalizability of our model, a necessary precondition to widespread application. DeepHeme demonstrated a high level of generalizability, evidenced by a decrease of only 4% in the mean F-1 score, from 0.89 to 0.85, across all 23 cell classes. Lastly, to improve access to the DeepHeme algorithm results and encourage further real-world generalizability testing, we developed a web application that allows scientists and clinicians to test the DeepHeme algorithm on either test images from our study or their own user-uploaded aspirates.

  PublisherDOI

• Fluctuation bounds for continuous time branching processes and evolution of growing trees with a change point

  The Annals of Applied Probability · 2023-07-10 · 6 citations

  articleSenior author

  We consider dynamic random trees constructed using an attachment function f:N→R+ where, at each step of the evolution, a new vertex attaches to an existing vertex v in the current tree with probability proportional to f(degree(v)). We explore the effect of a change point in the system; the dynamics are initially driven by a function f until the tree reaches size τ(n)∈(0,n), at which point the attachment function switches to another function, g, until the tree reaches size n. Two change point time scales are considered, namely the standard model where τ(n)=γn, and the quick big bang model where τ(n)=nγ, for some 0<γ<1. In the former case, we obtain deterministic approximations for the evolution of the empirical degree distribution (EDF) in sup-norm and use these to devise a provably consistent nonparametric estimator for the change point γ. In the latter case, we show that the effect of pre-change point dynamics asymptotically vanishes in the EDF, although this effect persists in functionals such as the maximal degree. Our proofs rely on embedding the discrete time tree dynamics in an associated (time) inhomogeneous continuous time branching process (CTBP). In the course of proving the above results, we develop novel mathematical techniques to analyze both homogeneous and inhomogeneous CTBPs and obtain rates of convergence for functionals of such processes, which are of independent interest.

  PublisherDOI

• Measure of Strength of Evidence for Visually Observed Differences between Subpopulations

  Journal of Computational and Graphical Statistics · 2023-11-02

  articleOpen access

  For measuring the strength of visually-observed subpopulation differences, the Population Difference Criterion is proposed to assess the statistical significance of visually observed subpopulation differences. It addresses the following challenges: in high-dimensional contexts, distributional models can be dubious; in high-signal contexts, conventional permutation tests give poor pairwise comparisons. We also make two other contributions: Based on a careful analysis we find that a balanced permutation approach is more powerful in high-signal contexts than conventional permutations. Another contribution is the quantification of uncertainty due to permutation variation via a bootstrap confidence interval. The practical usefulness of these ideas is illustrated in the comparison of subpopulations of modern cancer data.

  PublisherOA PDFDOI

Recent grants

• PostDoctoral Research Fellowship

  NSF · $150k · 2019–2023

Frequent coauthors

• Marina Gaeta Gazzola

  Bellevue Hospital Center

  13 shared

• Declan T. Barry

  Yale University

  11 shared

• Lynn M. Madden

  Yale University

  11 shared

• Xiaoying Zheng

  APT Foundation

  10 shared

• Mark Beitel

  7 shared

• Faisal Mahmood

  Broad Institute

  5 shared

• Drew F. K. Williamson

  5 shared

• Tiffany Chen

  Dartmouth College

  5 shared

Labs

• Computational Pathology GroupPI

  We are an interdisciplinary group working in computational Pathology. We build data driven, computational systems to analyze high-resolution histology images of diseased tissue and other clinical data sources.

Education

• Ph.D., Statistics

  University of North Carolina

• Other, Brigham and Women's Hospital/Harvard Medical School

  Division of Computational Pathology

• Other, University of Washington

  Department of Statistics

Awards & honors

• NSF Mathematical Sciences postdoctoral fellowship in the Dep…