Tavolara, Thomas Erol

Abstract Background Interpretation of multi-analyte laboratory profiles, such as lupus anticoagulant (LAC) testing, requires integration of numerous test results and clinical information. However, this process is time-consuming, requires domain knowledge, and has limited scalability. Given that rule-based systems are rigid and difficult to maintain and large language models pose risks of “hallucinations,” we developed and evaluated a novel retrieval-based approach to automate LAC profile interpretation to improve efficiency while maintaining fidelity and reliability for clinical use. Methods Our method processes laboratory values through ensemble random forest (RF) models to retrieve the most similar expert interpretation from a curated database of 347 previously interpreted LAC profiles. The system's performance was assessed through technical accuracy metrics, a work flow efficiency timing study, and expert review of interpretation accuracy for held-out cases. Results Technical validation demonstrated high accuracy in retrieving correct interpretative elements (median: 93% accuracy, interquartile range [IQR]: 67% to 100%). Retrieval accuracy was highly correlated with the prevalence of a given element (ρ = 0.8307). In a timing study of 93 consecutive LAC profiles, ensemble-assisted interpretation achieved a median time of 22 s per report (IQR: 16 s to 31 s), representing a 78.6% total reduction compared to manual interpretation (median: 118 s, IQR: 96 s to 172 s; t = 18.61, P < 0.001). Furthermore, 92.5% of AI-assisted interpretations were completed within 60 s, exceeding the target of 85%. Expert review confirmed 100% concordance for LAC presence/absence. Conclusions The proposed approach significantly improves efficiency of complex laboratory interpretation for LAC profiles, while maintaining consistency and high fidelity. Future studies will explore the method’s inherent extensibility to the interpretation of other multi-analyte laboratory profiles.

Abstract Objectives Traditional laboratory regulatory frameworks provide robust guidance for conventional clinical testing but lack requirements for data documentation for artificial intelligence and machine learning (AI/ML) solutions. This gap creates significant risks since AI solutions directly inherit patterns, biases, and limitations from their training data. Here, we describe the development and implementation of a comprehensive data documentation checklist with accompanying data lineage for use within an AI lifecycle framework for clinical laboratories. Methods Building on the previously established Clinical AI Readiness Evaluator (CARE) framework, we developed a comprehensive data checklist and data lineage methodology through a multiphase process, including (1) comprehensive review of existing AI/ML data documentation frameworks, (2) focused meetings with 3 institutional AI operations teams, and (3) 3 rounds of iterative refinement by our multidisciplinary team. The checklist’s effectiveness was then assessed using 3 diverse AI/ML projects moving through the CARE framework. Results The CARE Data Checklist and Data Lineage provide a structured approach to documenting critical aspects of datasets used in AI/ML projects, including data composition, demographics, collection methods, labeling processes, usage constraints, maintenance requirements, and a data readiness assessment. The checklist addresses unique data-centric challenges of AI/ML applications, facilitating transparency, reproducibility, and regulatory compliance. Conclusions The CARE Data Checklist and Data Lineage serve as both a technical guide and a communication tool bridging gaps between technical and clinical stakeholders. By working on these documents early in the AI lifecycle, laboratories can anticipate and address data-related challenges, ultimately saving time, optimizing resources, and improving the reliability of AI-augmented laboratory solutions.

Abstract Background: Recent advances in computer vision have given rise to image foundation models (FMs), large-scale neural networks pre-trained on millions of unlabeled histopathology patches, which can learn generalizable representations of tissue morphology. In many cancers, these FMs have predicted genetic mutations and prognoses directly from histology, suggesting that morphology encodes “hidden” molecular information. In lymphoma, this field is nascent but growing.General-purpose histopathology FMs have limited representation of lymphoid morphology. We developed and evaluated LymphoVision, a lymphoma-specialized image foundation model pretrained with self-supervised learning, to extract clinically meaningful features from routine hematoxylin and eosin (H&E) histology. Methods: LymphoVision was pre-trained using knowledge distillation and consistency (DINOv2) on over 37 million multi-resolution patches (5x, 10x, 20x, 40x) sampled using a density-aware strategy from 31,211 H&E whole slide images (WSIs) from 9,155 archival lymphoma and reactive lymphoproliferative cases. To ensure diverse, representative samples, our patch sampling method uses an efficient graph-based clustering algorithm and prioritizes morphologically informative regions by clustering patch embeddings and performing density sampling within each cluster. The pretrained encoder was evaluated using a clustering-constrained attention multiple instance learning (MIL) framework on three diagnostic tasks: (1) cell-of-origin (COO) classification in diffuse large B-cell lymphoma (DLBCL), distinguishing germinal center B-cell (GCB) from non-GCB subtypes based on the Hans immunohistochemistry algorithm; (2) follicular lymphoma (FL) grading, differentiating low-grade (grades 1–2) from high-grade (grade 3); and (3) multiclass lymphoma subtype classification across six categories: DLBCL, FL, marginal zone lymphoma (MZL), mantle cell lymphoma (MCL), classical Hodgkin lymphoma (CHL), and benign/reactive lymphoid conditions. WSI-level labels were derived from pathology reports, and none of the evaluation cases were included in pretraining. Models used a ViT-Giant LymphoVision backbone to generate 20x patch embeddings, followed by MIL using 50% training data and 2-fold cross-validation (25% validation, 25% test). Performance was assessed using area under the receiver operating characteristic curve (AUC), accuracy, recall (sensitivity) and F1 score. Results: For DLBCL COO classification (total n=358 cases), LymphoVision achieved a mean test AUC of 0.93 (range 0.92–0.94), accuracy of 84.5% and F1 of 0.85 across two independent test folds (n=91 per fold, 50% GCB). These results outperformed general-purpose FMs on the same task and cases, including Virchow-v2 (AUC 0.85) and UNI-v2 (AUC 0.88), despite those models being pretrained on 10–100 times more images. For FL grading (total n=660 cases), the test set in each fold included 166 cases (18% grade 3), yielding a mean test AUC of 0.88 (range 0.87–0.90), accuracy of 89.2% and F1 of 0.80. In multiclass lymphoma subtype classification (total n=1,176 cases), the test set included 296 cases per fold. LymphoVision achieved a mean test AUC of 0.98 (range 0.97-0.99), mean test accuracy of 87.1% and mean F1 of 0.84. Class-wise recall was 92.0% for DLBCL (n=112), 88.2% for FL (n=72), 77.2% for MZL (n=29), 70.8% for MCL (n=35), 87.5% for CHL (n=24), and 95.8% for benign/reactive lymphoid conditions (n=24). Conclusion: LymphoVision is a lymphoma-specialized FM trained using self-supervision that achieves high diagnostic accuracy across multiple clinically meaningful tasks using weak supervision and standard H&E images, suggesting the encoder has learned fine-grained lymphoma morphology. Based on historical benchmarks from the 1997 International Lymphoma Study Group classification project, expert hematopathologists achieved diagnostic agreement rates of 55% to 84% using H&E morphology alone for the non-Hodgkin lymphoma (NHL) entities included in our multiclass task. LymphoVision approached or exceeded these benchmarks. These findings underscore the value of our approach using disease-specialized pretraining and intelligent patch sampling for data-efficient histopathology learning. LymphoVision has the potential to augment diagnostics, enable scalable AI-powered classification in real-world workflows, and serve as a foundation for prognostic and theragnostic applications. External validation and multimodal integration are ongoing.

AbstractBreast cancer affects millions globally, necessitating precise biomarker testing for effective treatment. HER2 testing is crucial for guiding therapy, particularly with novel antibody‐drug conjugates (ADCs) like trastuzumab deruxtecan, which shows promise for breast cancers with low HER2 expression. Current HER2 testing methods, including immunohistochemistry (IHC) and in situ hybridization (ISH), have limitations. IHC, a semi‐quantitative assay, is prone to interobserver variability. While ISH provides higher precision than IHC, it remains more resource‐intensive in terms of cost and workflow. However, turnaround time is typically faster than that of other advanced molecular methods such as next‐generation sequencing. We adapted the clustering‐constrained‐attention multiple‐instance deep learning model to improve IHC testing and reduce dependence on reflex fluorescence ISH (FISH) tests. Using 5,731 HER2 IHC images, including 592 cases with FISH testing, we trained two models: one for predicting HER2 scores from IHC images and another for predicting FISH scores from equivocal cases. The HER2 IHC score prediction model achieved 91% ± 0.01 overall accuracy and a receiver operating characteristic (ROC) area under the curve (AUC) of 0.98 ± 0.01. The FISH score prediction model had an ROC AUC of 0.84 ± 0.07, with sensitivity at 0.37 ± 0.13 and specificity at 0.96 ± 0.03. External validation on cases from 203 institutions showed similar performance. The HER2 IHC model maintained a 91% ± 0.01 accuracy and an ROC AUC of 0.98 ± 0.01, while the FISH model had an ROC AUC of 0.75 ± 0.03, with sensitivity at 0.28 ± 0.04 and specificity at 0.93 ± 0.01. Our model advances HER2 scoring by reducing subjectivity and variability in current scoring methods. Despite lower accuracy and sensitivity in the FISH prediction model, it may be a beneficial option for settings where reflex FISH testing is unavailable or prohibitive. With high specificity, our model can serve as an effective screening tool, enhancing breast cancer diagnosis and treatment selection.

Abstract Histopathology is the gold standard for distinguishing tumor recurrence from treatment-induced necrosis but differentiating pseudoprogression from true recurrence is challenging in tumors with ambiguous morphological features. Prior research reported a 36% inter-rater agreement during suspected recurrence (Holdhoff et al., 2019). While serial radiographic and laboratory assessments are common, serial histopathological assessments are not. This study evaluated agreement and factors influencing assessment of active and inactive tumors through single and multiple time-point histopathological assessments. Patients who underwent initial and subsequent reoperation were identified, and their available histology slides were examined retrospectively. Four independent neuropathologists conducted two types of reviews: single time-point reviews using most recent biopsy/surgery slides, and multiple time-point reviews comparing images from initial and repeat biopsies taken 1-3 months apart. Neuropathologists completed a questionnaire assessing various tumor characteristics. Assignments were categorized as "active tumor," "treatment effect," or "unable to classify." 324 assessments were conducted (108 single time-point, 216 multiple time-point). Intraclass correlation coefficients for single vs. multiple time-point assessments were 0.72-0.93 for active tumors and 0.57-0.86 for inactive tumors. Pathologists' assessment of % inactive tumor was influenced by mitotic activity (p= 0.0028), cellularity (p< 0.0001), hyalinized vessels (p= 0.0019), and necrosis (p= 0.0046). Evaluation of % active tumor was influenced by hyalinized vessels (p= 0.0006), mitotic activity (p< 0.0001), cellularity (p< 0.0001), and microvascular proliferation (p= 0.0093), with the latter three factors specifically affecting final sign-out pathology. Serial histopathological assessment didn’t affect determination of % active or inactive tumor. No association between % inactive tumor and fibrinoid necrosis for single time-point (p= 0.34) but a strong negative association was seen for serial assessment (p= 0.0006). Inter-rater agreement was higher for identifying active tumors than inactive tumors in both case reviews. Changes in necrosis between tissue samples in single vs. multiple timepoint assessments influenced pathologists' assessment of % inactive tumor.

BACKGROUND: Surgical intervention remains the cornerstone of a multidisciplinary approach in the treatment of colorectal liver metastases (CLM). Nevertheless, patient outcomes vary greatly. While predictive tools can assist decision-making and patient counseling, decades of efforts have yet to result in generating a universally adopted tool in clinical practice. STUDY DESIGN: An international collaborative database of CLM patients who underwent surgical therapy between 2000 and 2018 was used to select 1,004 operations for this study. Two different machine learning methods were applied to construct 2 predictive models for recurrence and death, using 128 clinicopathologic variables: gradient-boosted trees (GBTs) and logistic regression with bootstrapping (LRB) in a leave-one-out cross-validation. RESULTS: Median survival after resection was 47.2 months, and disease-free survival was 19.0 months, with a median follow-up of 32.0 months in the cohort. Both models had good predictive power, with GBT demonstrating a superior performance in predicting overall survival (area under the receiver operating curve [AUC] 0.773, 95% CI 0.743 to 0.801 vs LRB: AUC 0.648, 95% CI 0.614 to 0.682) and recurrence (AUC 0.635, 95% CI 0.599 to 0.669 vs LRB: AUC 0.570, 95% CI 0.535 to 0.601). Similarly, better performances were observed predicting 3- and 5-year survival, as well as 3- and 5-year recurrence, with GBT methods generating higher AUCs. CONCLUSIONS: Machine learning provides powerful tools to create predictive models of survival and recurrence after surgery for CLM. The effectiveness of both machine learning models varies, but on most occasions, GBT outperforms LRB. Prospective validation of these models lays the groundwork to adopt them in clinical practice.