RANGER: Sparsely-Gated Mixture-of-Experts with Adaptive Retrieval Re-ranking for Pathology Report Generation

Yixin Chen; Ziyu Su; Hikmat Khan; Muhammad Khalid Khan Niazi

cs.CVMar 4, 2026

RANGER: Sparsely-Gated Mixture-of-Experts with Adaptive Retrieval Re-ranking for Pathology Report Generation

Authors: Yixin Chen, Ziyu Su, Hikmat Khan, Muhammad Khalid Khan Niazi

Organizations: Department of Biomedical Engineering, The Ohio State University, Columbus, OH, USA · Department of Pathology, The Ohio State University, Columbus, OH, USA

Abstract

Pathology report generation remains a relatively under-explored downstream task, primarily due to the gigapixel scale and complex morphological heterogeneity of Whole Slide Images (WSIs). Existing pathology report generation frameworks typically employ transformer architectures, relying on a homogeneous decoder architecture and static knowledge retrieval integration. Such architectures limit generative specialization and may introduce noisy external guidance during the report generation process. To address these limitations, we propose RANGER, a sparsely-gated Mixture-of-Experts (MoE) framework with adaptive retrieval re-ranking for pathology report generation. Specifically, we integrate a sparsely gated MoE into the decoder, along with noisy top- $k$ routing and load-balancing regularization, to enable dynamic expert specialization across various diagnostic patterns. Additionally, we introduce an adaptive retrieval re-ranking module that selectively refines retrieved memory from a knowledge base before integration, reducing noise and improving semantic alignment based on visual feature representations. We perform extensive experiments on the PathText-BRCA dataset and demonstrate consistent improvements over existing approaches across standard natural language generation metrics. Our full RANGER model achieves optimal performance on PathText dataset, reaching BLEU-1 to BLEU-4 scores of 0.4598, 0.3044, 0.2036, and 0.1435, respectively, with METEOR of 0.1883, and ROUGE-L of 0.3038, validating the effectiveness of dynamic expert routing and adaptive knowledge refinement for semantically grounded pathology report generation.

Figures & tables

Figure 1: Overview of the proposed RANGER framework. (a) Visual branch for whole-slide image feature extraction. (b) Dual learnable tokens that cross-attend to visual features and retrieved textual knowledge, respectively, enabling cross-modal interaction. (c) Construction of a sentence-level memory bank from historical pathology reports. (d) Two-stage retrieval module that first performs similarity-based candidate selection and then applies a learned re-ranking mechanism for adaptive knowledge refinement. (e) Sparsely-gated Mixture-of-Experts (MoE) decoder with noisy top- $k$ routing and load-balancing regularization for conditional expert specialization during report generation.

Model	BLEU-1 $\uparrow$	BLEU-2 $\uparrow$	BLEU-3 $\uparrow$	BLEU-4 $\uparrow$	METEOR $\uparrow$	ROUGE-L $\uparrow$
CNN-RNN [ 24 ]	0.371	0.185	0.089	0.043	0.143	0.239
att-LSTM [ 26 ]	0.371	0.191	0.094	0.048	0.142	0.238
Vanilla Transformer [ 23 ]	0.389	0.246	0.157	0.103	0.158	0.257
R2Gen [ 8 ]	0.378	0.243	0.160	0.107	0.179	0.279
R2GenCMN [ 7 ]	0.396	0.254	0.164	0.110	0.163	0.279
MI-Gen [ 4 ]	0.416	0.267	0.174	0.115	0.165	0.270

Table 1: Results of pathology report generation on PathText (BRCA). BLEU-n denotes n-gram BLEU score.

\uparrow

indicates higher is better. Bold denotes the best result.

Figure 2: Qualitative comparison of generated and reference pathology report text. Correctly generated words and phrases are highlighted in red .

Model	Reranker	MoE	Noisy Top- $k$	Load Balance	BLEU-1 $\uparrow$	BLEU-2 $\uparrow$	BLEU-3 $\uparrow$	BLEU-4 $\uparrow$	METEOR $\uparrow$	ROUGE-L $\uparrow$
Baseline (cosine retrieval)	$\times$	$\times$	$\times$	$\times$	0.4500	0.2960	0.1960	0.1350	0.1800	0.2930
+ MLP Reranker	$\checkmark$	$\times$	$\times$	$\times$	0.4543	0.2997	0.1993	0.1387	0.1837	0.2978
+ MoE Decoder (sparse)	$\checkmark$	$\checkmark$	$\times$	$\times$	0.4540	0.2989	0.1987	0.1380	0.1829	0.3005
+ Noisy Top- $k$ Routing	$\checkmark$	$\checkmark$	$\checkmark$	$\times$	0.4569	0.3019	0.2013	0.1410	0.1858	0.3022
+ Load Balance ( $\lambda=0.01$ ) (Ours)	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	0.4598	0.3044	0.2036	0.1435	0.1883	0.3038

Table 2: Ablation study on PathText (BRCA). BLEU-n denotes n-gram BLEU score.

\uparrow

indicates higher is better.

Recall Size $K$	BLEU-1	BLEU-2	BLEU-3	BLEU-4	METEOR	ROUGE-L
$K=10$	0.4568	0.3022	0.2018	0.1420	0.1868	0.3018
$K=20$ (Ours)	0.4598	0.3044	0.2036	0.1435	0.1883	0.3038
$K=50$	0.4576	0.3028	0.2023	0.1426	0.1873	0.3025

Table 3: Stage-1 recall size

K

(top-

k

=3 fixed).

Top- $k$	BLEU-1	BLEU-2	BLEU-3	BLEU-4	METEOR	ROUGE-L
$k=1$	0.4561	0.3016	0.2012	0.1415	0.1862	0.3014
$k=3$ (Ours)	0.4598	0.3044	0.2036	0.1435	0.1883	0.3038
$k=5$	0.4580	0.3031	0.2025	0.1428	0.1875	0.3029

Table 4: Final top-

k

selection (

K=20

fixed).

Figure 3: Ablation study on the Mixture-of-Experts (MoE) decoder. Left: Performance variation with different numbers of experts. Right: Effect of top- $k$ routing during expert activation. Moderate expert capacity with sparse routing achieves the best trade-off between specialization and stability.

Figure 4: Effect of load-balance coefficient $\lambda$ on MoE decoder performance (BLEU-4). $\lambda$ =0.01 yields the best performance. Too-small $\lambda$ (0.001) fails to prevent expert collapse and performs worse than no load balancing, while too-large $\lambda$ (0.1) over-constrains routing. Dashed line: cosine-retrieval baseline

Medical Report Generation

Medical report generation turns medical images, clinical records, or conversations into structured, clinically useful reports.

See all 82 papers

PathReportEval: A Systematic Benchmark for Pathology Report Generation
Jul 20, 2026Suryakant Singh, Sejuti Majumder, Beatrice Knudsen +2Medical Report Generation Medical Image Benchmarks
Department of Biomedical Informatics Stony Brook University Stony Brook, NY, USA · Department of Pathology University of Utah Salt Lake City, UT, USA
Semantic Context-aware mOdality fUsion Transformer (SCOUT): A Context-Aware Multimodal Transformer for Concept-Grounded Pathology Report Generation
May 1, 2026Suryakant Singh, Saarthak Kapse, Joel Saltz +1Computational Pathology Medical Report Generation
Stony Brook University
Simple Token-Efficient Vision-Language Model for Case-level Pathology Synoptic Report Generation
May 29, 2026Zhiyuan Yang, Jiahao Cheng, Vincent Quoc-Huy Trinh +1Computational Pathology Medical Report Generation
Department of Computer Science and Software Engineering (CSSE), Concordia University, Montreal, Canada · Axe Cancer, Centre de recherche du CHUM, Université de Montréal, Montreal, Canada · Institut de recherche en immunologie et cancérologie (IRIC), Université de Montréal

Jul 20, 2026J/K move · Enter open · S save

PathReportEval: A Systematic Benchmark for Pathology Report Generation

Suryakant Singh, Sejuti Majumder, Beatrice Knudsen, Joel Saltz+1

Department of Biomedical Informatics Stony Brook University Stony Brook, NY, USA · Department of Pathology University of Utah Salt Lake City, UT, USA

Medical Report Generation Medical Image Benchmarks Whole-Slide Image Analysis

Open paper AskPDF arXiv

Pathology report generation from whole-slide images (WSIs) is a rapidly growing multimodal learning problem, yet progress is difficult to measure because existing studies use heterogeneous datasets, model settings, visual encoders, and evaluation protocols. Moreover, commonly used natural language generation metrics, including BLEU, ROUGE, and METEOR, primarily reward lexical similarity and often fail to detect clinically consequential errors such as omitted diagnoses, hallucinated findings, or discordant tumor attributes. We present a standardized benchmark and evaluation framework for pathology report generation. The benchmark evaluates four representative methods across three datasets (TCGA, HistAI, and REG 2025) using three pathology foundation encoders (CONCHv1.5, UNI2-h, and H-Optimus-1). Our framework standardizes preprocessing, feature extraction, training, decoding, and evaluation, enabling fair comparison across models while providing a modular platform for integrating new methods, datasets, and encoders. A central contribution is the Clinical Report Quality Score (CRQS), a clinically grounded metric for evaluating factual correctness. CRQS maps reference and generated reports into structured clinical attributes and measures four complementary dimensions: clinical fact coverage, key information recall, hallucination rate, and clinical discordance, producing both an overall score and interpretable sub-scores. Experiments demonstrate that conventional language-generation metrics are weakly aligned with clinical correctness and frequently overestimate report quality. In contrast, CRQS reveals clinically meaningful differences between models and encoders that lexical metrics fail to capture. Together, the benchmark, public plug-and-play framework, and CRQS establish a reproducible foundation for rigorous evaluation of pathology report generation.

Abstract

Medical Report Generation

Explore similar work

PathReportEval: A Systematic Benchmark for Pathology Report Generation

Semantic Context-aware mOdality fUsion Transformer (SCOUT): A Context-Aware Multimodal Transformer for Concept-Grounded Pathology Report Generation

Simple Token-Efficient Vision-Language Model for Case-level Pathology Synoptic Report Generation