FD-AA: A Lightweight Focal-Diffuse And Attenuation-Aware Head for Incidental Abdominal Abnormality Detection in Chest CT

Haoyan Ding; Kritika Iyer; Halid Yerebakan; Zhenyu Bu; Chushu Shen; Peiyu Duan; Xinyuan Zheng; Sepehr Farhand

cs.CVSep 28, 2026

FD-AA: A Lightweight Focal-Diffuse And Attenuation-Aware Head for Incidental Abdominal Abnormality Detection in Chest CT

Authors: Haoyan Ding, Kritika Iyer, Halid Yerebakan, Zhenyu Bu, Chushu Shen, Peiyu Duan, Xinyuan Zheng, Sepehr Farhand, +4 more

Organizations: Siemens Medical Solutions USA, Inc., Malvern, PA, USA · University of California, Los Angeles, Los Angeles, CA, USA

Abstract

Routine chest CT captures upper-abdominal structures that may contain clinically relevant incidental abnormalities. Detecting these findings requires feature extraction from organs with different spatial extents and attenuation patterns. We propose FD-AA, a lightweight organ-aware classification head adaptable for frozen 3-D CT encoders. Within each organ, an attenuation-aware module preserves sparse focal evidence, while masked generalized-mean pooling captures diffuse anomaly patterns. In seven abdominal organs, FD-AA with Pillar-0 achieved state-of-the-art (SOTA) performance in both the CT-RATE test set (AUC = 0.798) and the external RAD-ChestCT dataset (AUC = 0.713). More specifically, FD-AA improved macro AUC/AP from 0.763/0.346 to 0.798/0.405 over direct classification using frozen Pillar-0 only (p = 0.034/0.016). Such performance gain generalizes across multiple frozen encoders (AUC improvement on MedicalNet +9.8%, CT-CLIP +14.7%, ResNet +3.7%), demonstrating the effectiveness of FD-AA across different feature representations. These results support the effectiveness of integrating focal-diffuse aggregation with explicit HU evidence for incidental abdominal abnormality detection.

Figures & tables

Figure 1: Overview of FD-AA. A frozen Pillar-0 encoder extracts spatial tokens and a global embedding from the input chest CT, while TotalSegmentator supplies organ masks. For each organ, focal pathway uses Top- $k$ pooling to retain sparse focal evidence, whereas diffuse pathway uses GeM to summarize spatially distributed changes. Both representations are combined with global context. Besides, organ-masked HU maps encode high- and low-attenuation evidence while discounting adjacent intensity confounders. A lightweight adapter injects this feature as a residual correction to the focal pathway.

Figure 2: Representative organ-aware attention visualizations. Each case is shown as an unannotated CT image (left) and a token-level attention overlay with mask (right). Green contours denote the TotalSegmentator organ masks, and the colored rectangular patches show token attention, with warmer colors indicating greater weight.

		CT-RATE (internal test)				RAD-ChestCT (external)
	Method	AUC $\uparrow$	AP $\uparrow$	F1 $\uparrow$	Acc. $\uparrow$	AUC $\uparrow$	AP $\uparrow$	F1 $\uparrow$	Acc. $\uparrow$
Global	COLIPRI-CRM [ 23 ]	0.761	0.357	0.415	0.579	0.644	0.227	0.285	0.310
	RadFinder [ 10 ]	0.711	0.304	0.440	0.839	0.612	0.218	0.316	0.823
	ItemizedCLIP [ 17 ]	0.708	0.305	0.384	0.520	0.545	0.170	0.258	0.182
	Percival [ 3 ]	0.642	0.271	0.224	0.385	0.526	0.153	0.193	0.713
	Merlin [ 4 ]	0.585	0.226	0.289	0.242	0.523	0.167	0.235	0.414

Table 1: Comparison with 3-D CT foundation-model references over seven abdominal targets. CT-RATE is the internal test cohort (

n=3{,}039

), and RAD-ChestCT is the external cohort (

n=3{,}630

). Bold and underlining mark the best and second-best result in each column, respectively. Groups reflect anatomical modeling.

Encoder	Method	AUC	AP	F1	Prec.	Rec.	Acc.	Any-AUC
Pillar-0	GA	0.763	0.346	0.428	0.348	0.557	0.835	0.805
	OA	0.768	0.355	0.461	0.389	0.568	0.853	0.809
	FD	0.790	0.395	0.514	0.449	0.602	0.874	0.826
	FD-AA	0.798	0.405	0.519	0.461	0.594	0.878	0.838
MedicalNet	GA	0.682	0.213	0.296	0.198	0.587	0.690	0.733
	OA	0.695	0.250	0.350	0.254	0.572	0.765	0.762

Table 2: Evaluation Across Image Encoders on CT-RATE. Rows within each encoder compare classification heads.

Explore similar work

EXACT: an explainable anomaly-aware vision foundation model for analysis of 3D chest CT
Apr 27, 2026Xuguang Bai, Mingxuan Liu, Tongxi Song +6Cone-Beam Computed Tomography Recent Vision Foundation Models
School of Biomedical Engineering, Tsinghua University, Beijing, China. · Department of Radiology, Mianyang Central Hospital, Mianyang, China. · Center for Biomedical Imaging Research, Tsinghua University, 30 Shuangqing Road, Haidian District, Beijing, China, 100084
CT-IDP: Segmentation-Derived Quantitative Phenotypes for Interpretable Abdominal CT Disease Classification
May 9, 2026Lavsen Dahal, Joseph Y. LoPhenotypes Disease
1. Center for Virtual Imaging Trials, Carl E. Ravin Advanced Imaging Laboratories · 3. Electrical and Computer Engineering, Pratt School of Engineering, Duke University · 4. Medical Physics Graduate Program, Duke University +1
ORACLE-CT: Anatomy-Aware Support Pooling for CT Classification
Jun 3, 2026Lavsen Dahal, Yubraj Bhandari, Geoffrey Rubin +1Semi-Supervised Medical Image Segmentation Global Average Pooling
Electrical and Computer Engineering, Pratt School of Engineering, Duke University, Durham, 27708, North Carolina, USA · Department of Mathematics, Trinity College of Arts & Sciences, Duke University, Durham, 27708, North Carolina, USA · Department of Radiology and Imaging Sciences, University of Arizona College of Medicine, Tucson, 85004 Arizona, USA +1

Apr 27, 2026, cs.CVJ/K move · Enter open · S save

EXACT: an explainable anomaly-aware vision foundation model for analysis of 3D chest CT

Xuguang Bai, Mingxuan Liu, Tongxi Song, Yifei Chen+5

School of Biomedical Engineering, Tsinghua University, Beijing, China. · Department of Radiology, Mianyang Central Hospital, Mianyang, China. · Center for Biomedical Imaging Research, Tsinghua University, 30 Shuangqing Road, Haidian District, Beijing, China, 100084

Cone-Beam Computed Tomography Recent Vision Foundation Models Voxel Foundation Model Report Exact

Open paper AskPDF arXiv

Chest computed tomography (CT) is central to the detection and management of thoracic disease, yet the growing scale and complexity of volumetric imaging increasingly exceed what can be addressed by scan-level prediction alone. Clinically useful AI for CT must not only recognize disease across the whole volume, but also localize abnormalities and provide interpretable visual evidence. Existing vision-language foundation models typically compress scans and reports into global image-text representations, limiting their ability to preserve spatial evidence and support clinically meaningful interpretation. Here we developed EXACT, an explainable anomaly-aware foundation model for three-dimensional chest CT that learns spatially resolved representations from paired clinical scans and radiology reports. EXACT was pre-trained on 25,692 CT-reports pairs using anatomy-aware weak supervision, jointly learning organ segmentation and multi-instance anomaly localization without manual voxel-level annotations. The resulting organ-specific anomaly-aware maps assign each voxel a disease-specific anomaly score confined to its corresponding anatomy, jointly encoding lesion extent and organ-level context. In retrospective multinational and multi-center evaluations, EXACT showed broad and consistent improvements across clinically relevant CT tasks, spanning multi-disease diagnosis, zero-shot anomaly localization, downstream adaptation, and visually grounded report generation, outperforming existing three-dimensional medical foundation models. By transforming routine clinical CT scans and free-text reports into explainable voxel-level representations, EXACT establishes a scalable paradigm for trustworthy volumetric medical AI.

Abstract

Explore similar work

EXACT: an explainable anomaly-aware vision foundation model for analysis of 3D chest CT

CT-IDP: Segmentation-Derived Quantitative Phenotypes for Interpretable Abdominal CT Disease Classification

ORACLE-CT: Anatomy-Aware Support Pooling for CT Classification