Tools & Data

HistoQC provides automated quality assessment for whole-slide pathology images, detecting artifacts including blur, tissue fold, air bubbles, pen marks, and staining anomalies. The tool generates per-tile quality masks and summary metrics, enabling downstream analysis to filter unreliable regions automatically.

Janowczyk A, et al. JCO CCI 2019.

F-SYN performs stain normalization in the Fourier domain, transferring stain statistics between source and target slides while preserving tissue morphology. Unlike GAN-based approaches, F-SYN introduces no hallucinated structures, making it suitable for clinical and regulatory workflows.

MQUAL provides automated quality scoring for prostate and breast MRI studies, evaluating signal-to-noise ratio, motion artifact severity, field uniformity, and sequence protocol compliance. Validated on EPIP and VA prostate MRI datasets, MQUAL enables scalable QC for large retrospective imaging cohorts.

Beaks provides a unified quality assurance interface for heterogeneous imaging collections, supporting both radiology (DICOM) and pathology (SVS/NDPI) formats. It generates standardized quality reports and flags outlier studies for radiologist or pathologist review.

PyComBatch implements ComBat harmonization with extensions for radiomic feature matrices, removing scanner-related batch effects while preserving biological variance. Supports both parametric and non-parametric harmonization modes, with integration hooks for IBSI-compliant feature sets.

APIC uses deep learning on H&E pathology slides to quantify tumor-immune microenvironment features and predict treatment benefit from systemic therapies. Validated on CHAARTED and STAMPEDE clinical trial datasets, APIC achieves performance comparable to expensive molecular assays at a fraction of the cost.

Bhatt D, et al. JCO CCI 2024.

MOSCARD addresses spurious correlations in multimodal fusion models by embedding causal structure into the learning process. Using structural causal models (SCMs) and do-calculus, MOSCARD separates true predictive features from confounders introduced by demographic or acquisition biases. Demonstrated on cardiovascular opportunistic screening with strong fairness properties.

Gichoya JW, et al. MICCAI 2025.

MamoCLIP adapts the CLIP contrastive learning paradigm to mammography analysis, aligning imaging features with radiology report text embeddings. Trained in a federated setting across EMBED v2 institutions, MamoCLIP produces transferable representations for downstream breast cancer classification and risk prediction tasks.

Shrivastava A, et al. MICCAI 2024.

HemaToMe provides a general-purpose multimodal fusion architecture combining image-derived pathology features with structured EHR data. The framework implements graph-based and attention-based fusion strategies with support for missing modality imputation, making it suitable for real-world clinical cohorts with incomplete data.

ProstateNet provides pixel-level segmentation of the transition zone (TZ) and peripheral zone (PZ) in biparametric prostate MRI, enabling zone-specific radiomic feature extraction aligned with PI-RADS v2.1. Trained on EPIP and VA datasets, ProstateNet generalizes robustly across scanner vendors and field strengths.