The starting point is the Segment Anything Model, or SAM, the vision foundation model released by Meta AI that learned to segment arbitrary objects from natural images at unprecedented scale. SAM’s encoder encodes rich, general-purpose visual representations, which is precisely why researchers have been eager to adapt it to medicine. But the model’s sheer size makes conventional fine-tuning impractical in clinical settings. Updating every parameter requires gigabytes of memory, and in a federated learning setup, where hospitals train locally and share model updates rather than data, transmitting full model weights between institutions would be prohibitively expensive and would expose the intellectual property embedded in the model itself.

FedMed-LoRA’s answer is a dual low-rank adaptation strategy. Low-rank adaptation, or LoRA, rests on a simple observation: the changes a network needs to learn when moving from one domain to another often live in a much smaller subspace than the full weight matrices. Instead of updating a weight matrix W directly, LoRA freezes W and learns two much smaller matrices whose product approximates the update. Because the rank of these small matrices is far lower than the dimensions of the original weights, the number of trainable parameters collapses by orders of magnitude while much of the adaptation capacity is preserved.

What distinguishes FedMed-LoRA is where it places those low-rank modules. The framework inserts adaptation pairs into both the attention pathways and the multilayer perceptron pathways of SAM’s encoder, which remains fully frozen throughout training. Attention layers govern how different parts of an image inform one another, while MLP layers transform individual features; the authors report that ablation experiments confirm the two LoRA paths play complementary roles, and that removing either degrades performance. This dual placement matters for medical segmentation because the task demands both global context, for instance understanding where an organ sits relative to its neighbors, and fine local discrimination of tissue boundaries that differ subtly from surrounding anatomy.

The second pillar of the framework is aggressive memory engineering. The frozen SAM weights are quantized to INT4, a four-bit integer representation that shrinks the memory footprint of the backbone dramatically compared with standard sixteen-bit floating point storage. Meanwhile, the small trainable low-rank modules are trained in BF16, a floating point format that preserves numerical stability during gradient updates. The combination means the bulk of the model occupies minimal memory while the parts that actually learn retain the precision they need. According to the authors, this design allows large foundation models to be fine-tuned on hospital-level hardware, the kind of commodity equipment a radiology department might realistically own rather than a dedicated GPU cluster.

Privacy is handled through the federated workflow itself. Each participating institution trains the dual LoRA modules on its own scans, and only those low-rank updates are shared and aggregated into a global model. Patient images never leave the hospital, and because the frozen backbone is never transmitted, the underlying model’s intellectual property stays protected as well. This addresses a persistent obstacle in multi-institutional medical AI: cross-center imaging heterogeneity, where different scanners, acquisition protocols, and patient populations cause models trained at one center to falter at another. Federated averaging over diverse sites is a natural way to build robustness, but only if the communication burden per round is small enough to be practical, which is exactly what parameter-efficient fine-tuning delivers.

The experimental results reported in the paper are striking. On the FLARE22 multi-organ segmentation benchmark, FedMed-LoRA matches the performance of full-parameter fine-tuning, meaning the tiny set of trainable low-rank parameters recovers essentially all of the adaptation benefit of updating hundreds of millions of weights. On the Synapse benchmark, which tests cross-domain generalization, the framework outperforms established medical adaptation baselines including MedSAM, Med-SA, and SAMed. That cross-domain result is arguably the more clinically meaningful one, since a model that only works on the data distribution it was tuned on is of limited use in the varied imaging environments of real hospitals.

The broader significance extends beyond one benchmark. Foundation models have transformed natural language processing and are reshaping computer vision, but medicine has lagged in adopting them, partly because the economics do not work: most hospitals cannot afford the hardware or the data-sharing agreements that naive fine-tuning would require. Parameter-efficient fine-tuning combined with quantization changes that calculus, reducing both the compute needed to adapt a model and the bandwidth needed to collaborate on that adaptation. If the pattern holds across other medical tasks, the same recipe of frozen quantized backbones plus lightweight trainable adapters could become a standard template for bringing frontier-scale models into regulated, privacy-sensitive domains.

There are, of course, caveats worth keeping in view. The framework was evaluated on public benchmarks rather than deployed in live clinical workflows, and the paper does not claim regulatory clearance or prospective validation. Quantization to four bits, while memory-efficient, can in principle introduce approximation errors, and the authors’ results suggest these are tolerable for segmentation but the trade-offs for other modalities remain to be demonstrated. Federated systems also face practical realities, from sites dropping out of training rounds to statistical heterogeneity that can bias aggregated updates, that benchmarks only partially capture. Still, the work offers a concrete, tested answer to a question that has shadowed medical AI for years: how to harness the representational power of enormous foundation models under the constraints that medicine actually imposes.

For now, FedMed-LoRA stands as a demonstration that the path from a general-purpose segmentation model to a privacy-preserving, hospital-deployable medical tool may be far shorter, and far lighter, than anyone assumed. By keeping the giant frozen, training only a sliver of parameters, and letting hospitals collaborate without sharing a single scan, the researchers have sketched what clinically deployable foundation-model adaptation could look like. The work was supported by the High-Level Talent Research Start-Up Fund of Shandong Xiehe University, and the article is open access, allowing other teams to scrutinize, reproduce, and extend the approach as the field moves from benchmarks toward the clinic.

Subject of Research: Federated parameter-efficient fine-tuning of vision foundation models for privacy-preserving medical multi-organ segmentation

Article Title: FedMed-LoRA: a federated parameter-efficient fine-tuning framework with dual lora for medical multi-organ segmentation

Article References: Zhou, F., Jiao, S., Lv, X., Zhou, N., Hu, H., Wang, G., Song, J., & Zhang, Z. (2026). FedMed-LoRA: a federated parameter-efficient fine-tuning framework with dual lora for medical multi-organ segmentation. Complex & Intelligent Systems. https://doi.org/10.1007/s40747-026-02548-1

Image Credits: AI Generated

DOI: 10.1007/s40747-026-02548-1

Keywords: FedMed-LoRA, federated learning, SAM, low-rank adaptation, parameter-efficient fine-tuning, multi-organ segmentation, medical imaging, quantization, privacy-preserving AI, foundation models, deep learning, Complex & Intelligent Systems

The problem the team set out to solve is deceptively simple to state and notoriously hard to execute. Accurate, objective, and scalable disease assessment underpins everything from fungicide timing to breeding decisions, but manual scouting cannot deliver it. Human evaluators differ from one another and from themselves over time; rater fatigue and variable lighting degrade consistency further; and coarse ordinal scales lack the resolution to capture subtle treatment effects. These inconsistencies ripple directly into genetics, where phenotype-to-genotype analyses depend on reliable measurements. Even when quantitative trait loci exist, noisy phenotypes can obscure them, reducing reproducibility across breeding programs. In regions like California, scouts must also peer into shaded inner canopy tissues where symptoms hide or pathogen viability is suppressed by ultraviolet light and heat.

Automated imaging systems have long promised a way out, capturing high-throughput visual data across entire vineyard blocks. But a bottleneck has persisted on the annotation side. Powdery mildew lesions are irregular, diffuse, and visually variable, making pixel-level labeling labor-intensive and dependent on scarce expert knowledge. Building large, high-quality, domain-specific datasets is prohibitively expensive, and conventional deep learning models trained from scratch on limited labeled data tend to falter in messy outdoor environments. The challenge, as the researchers frame it, is not acquiring imagery but teaching models to learn from scarce and noisy labels.

Their answer draws on the current paradigm shift in artificial intelligence from task-specific networks to generalist foundation models. The team fused two of the most influential vision foundation models: the Segment Anything Model, or SAM, trained on more than a billion masks and capable of promptable segmentation of arbitrary objects, and CLIP, a model trained on 400 million image-text pairs that aligns visual and linguistic representations in a shared embedding space. SAM contributes high-resolution spatial precision, preserving object boundaries and fine structure, while CLIP contributes semantic reasoning grounded in natural language. In the integrated architecture, each image is processed in parallel by both encoders. A learnable multilayer perceptron projects SAM’s embedding into CLIP’s latent space, the two representations are fused by element-wise addition, and the result is conditioned on text embeddings for prompts such as “Powdery mildew” or “Canopy.” A dot-product interaction between the fused visual representation and the text embedding produces a multi-modal representation that a second MLP transforms back into the spatial layout SAM’s mask decoder needs. The output is segmentation guided jointly by pixel-level detail and language-based disease concepts.

Because labeled agricultural data is scarce, the researchers systematically compared parameter-efficient fine-tuning strategies: full fine-tuning, adapter modules inserted into key layers, and low-rank adaptation, or LoRA, which updates attention and feedforward layers through small low-rank matrices. Each strategy was tested with two SAM backbones, ViT-B and ViT-H, and with either the encoder alone or both encoder and decoder trainable. The verdict was clear: adapter-based tuning with both modules trainable performed best, and the fine-tuning strategy mattered more than raw model size. LoRA’s tiny updates proved insufficient to transfer CLIP’s rich semantic knowledge into the agricultural domain, a finding that underscores how much representational capacity is needed to absorb new visual vocabulary.

The resulting models, PM-SAM-CLIP for lesion segmentation and Canopy-SAM-CLIP for canopy extraction, were trained on modest datasets collected in New York with custom imaging rigs using active strobe illumination to maximize contrast and consistency. PM-SAM-CLIP, built on the ViT-H backbone, was fine-tuned on just 175 manually annotated images; Canopy-SAM-CLIP used 415. Despite the small training sets, the models decisively outperformed baselines including PM-SAM, DeepLab v3+, and SegFormer. On the powdery mildew test set, PM-SAM-CLIP achieved 59.63 percent image-level mean intersection over union and 64.12 percent dataset-level, beating the strongest baseline by roughly 13 to 14 percentage points and weaker baselines by more than 30. Crucially, on the worst-performing subsets of images, where competing models nearly collapsed to zero, the multi-modal model maintained scores between 21.51 and 30.25 percent, demonstrating that semantic context stabilizes performance on the hardest cases, such as sparse lesions and cluttered canopies. Canopy segmentation reached 95.59 percent image-level accuracy with strong worst-case robustness.

The real test came in California, at a USDA vineyard in Parlier planted with a genetically diverse mapping population spanning multiple Vitis species and managed under a different trellis system than the New York training data. Deployed zero-shot, without any additional fine-tuning, PM-SAM-CLIP segmented infections across two growing seasons and more than 30,000 images, contending with confounders like dust and sunburn that visually mimic mildew. A post-processing step intersecting predicted infection masks with canopy masks suppressed false positives outside biologically relevant tissue. At the vine level, severity was computed as the maximum normalized overlap between infection and canopy masks across all images of each vine, mirroring how expert scouts assess the worst infection instance. The agreement with human experts was striking: 79 percent of vines in 2023 and 81 percent in 2024 received identical scores, and more than 98 percent fell within one score level of the human rating. The pipeline even captured subtleties humans cannot, such as the difference between 75 and 76 percent infected area.

The most consequential validation came from genetics. The team fed image-derived severity scores into QTL mapping using a high-density rhAmpSeq genetic map built from 465 markers across 291 progeny of a cross between Vitis amurensis and V. vinifera. Across both years, every phenotyping method, human and automated alike, converged on the same significant QTL on chromosome 13, spanning 22.2 to 27.6 megabases. That interval overlaps the previously fine-mapped REN12 locus, a known major resistance gene, providing biological confirmation that the AI-generated phenotypes capture real genetic signal. The image-derived QTL explained 22.9 to 26.2 percent of phenotypic variance, slightly below the human-scored range but well within expected field variability. Spatiotemporal heatmaps of infection across the vineyard additionally revealed persistent high-severity zones and stable low-scoring vines, information that can guide both genotype selection and site-specific fungicide decisions.

The implications extend well beyond one disease or one crop. By fine-tuning pre-trained foundation models on tiny domain datasets, the approach sidesteps the annotation bottleneck that has long constrained AI in agriculture, and text prompts give plant pathologists an interpretable way to steer model behavior using their own vocabulary. The researchers acknowledge limitations, including occasional false positives from dust and sunburn, QR-code detection failures, and the expressiveness limits of single-keyword prompts, and they point toward edge deployment on field robots, richer textual descriptions, and integration with hyperspectral or thermal sensing. They also call for open, FAIR-compliant multi-modal agricultural datasets to accelerate foundation-model training across the plant sciences. With code, models, and datasets publicly released, the study sketches a closed loop from field imagery to genotype discovery to precision disease management, a blueprint for next-generation phenotyping in viticulture and beyond.

Subject of Research: Automated phenotyping of grapevine powdery mildew using integrated SAM-CLIP multi-modal vision foundation models

Article Title: Integrating large multi-modal models for automated powdery mildew phenotyping in grapevines

Article References: Lin, Y., Dashner, Z., Jimenez, A., Wilkerson, D., Cadle-Davidson, L., Riaz, S., & Jiang, Y. (2026). Integrating large multi-modal models for automated powdery mildew phenotyping in grapevines. Artificial Intelligence in Agriculture. https://doi.org/10.1016/j.aiia.2026.09.006

Image Credits: AI Generated

DOI: 10.1016/j.aiia.2026.09.006

Keywords: powdery mildew, grapevine, phenotyping, SAM, CLIP, foundation models, segmentation, QTL mapping, viticulture, plant disease detection, multi-modal AI, breeding

The breadth of the AI4US collection becomes clearer when the individual contributions are grouped by clinical problem rather than by technique. A striking number of the papers converge on the heart. Echocardiography, the ultrasound-based assessment of cardiac structure and function, is among the most widely performed imaging examinations in the world, yet it remains one of the most demanding to interpret. Measurements such as left ventricular ejection fraction, a key indicator of how strongly the heart pumps, depend on tracing the boundaries of the left ventricle across many frames of a moving image sequence. Manual tracing is slow and varies between operators, which is why several studies in the issue tackle this exact task. One adapts the Segment Anything Model, a general-purpose vision foundation model, to echocardiographic sequences through an implicit decoding strategy; another integrates multi-scale information and diverse prompts into a large medical model to estimate ejection fraction directly; a third introduces a diffusion-based segmentation model, reflecting a broader trend in which generative modeling frameworks are repurposed for precise anatomical delineation.

Temporal information is a recurring theme in these cardiac works. A heart is a moving target, and a segmentation that ignores motion is throwing away much of the physics of the examination. The special issue includes a motion-enhanced, shape-aware semi-supervised network for echocardiographic sequence segmentation, as well as a semi-supervised approach to automatic four-dimensional mitral valve segmentation from transesophageal echocardiography. The latter is notable on two fronts. Transesophageal imaging places the probe close to the heart, yielding unusually detailed views, but four-dimensional datasets from this modality are enormous and laborious to annotate. Semi-supervised learning, in which a model learns from a small set of fully labeled examples alongside a much larger pool of unlabeled data, offers a pragmatic path forward, and its application to a structure as geometrically intricate as the mitral valve illustrates how far these methods have matured.

Beyond segmentation, the cardiac papers extend into prediction. One contribution addresses subclinical atrial fibrillation, a rhythm disturbance that often goes undetected because it comes and goes unpredictably, by combining deep learning with strain analysis of echocardiographic images. Strain refers to the deformation of heart muscle during the cardiac cycle, and subtle abnormalities in deformation can betray electrical disease before it declares itself on an electrocardiogram. Another cardiac study tackles view recognition, the task of determining which standard plane the sonographer is currently imaging, using an efficient network built on a state space model and trained with evidential learning, an approach that allows the network to express uncertainty about its own predictions. Standard plane recognition is a deceptively important problem: fetal echocardiography in particular requires the operator to find specific anatomical planes among dozens of possible positions, and automated recognition could serve both as a quality-control aid and as a training tool.

The fetus features prominently elsewhere in the collection as well. A study on operationalizing machine learning models for standard fetal ultrasound plane classification moves beyond model development into the engineering discipline of MLOps, the practice of deploying, monitoring, and maintaining machine learning systems in production. This shift in emphasis is significant. A model that performs well in a laboratory setting may fail silently in a clinic when image quality drifts, when hardware changes, or when patient populations differ from the training data. Operational concerns like these are increasingly recognized as the true frontier of clinical artificial intelligence, and their appearance in a special issue of this kind signals that the field is beginning to treat deployment as seriously as architecture design.

Obstetric imaging also demands extreme computational efficiency, because modern ultrasound systems can acquire volumetric data at high frame rates. One paper in the issue addresses deep learning-based high-precision three-dimensional ultrasound imaging for large organs, while another applies an efficient network with a state space model to fetal echocardiographic standard view recognition. State space models have recently emerged as an alternative to transformers for sequence modeling, offering the ability to capture long-range dependencies with computational cost that scales more favorably with input length. Their adoption in ultrasound research, where video and volume data strain the budgets of attention-based architectures, exemplifies how the field borrows and adapts the newest tools from the wider machine learning community.

Abdominal and vascular imaging form another major cluster. The liver is a frequent target: one study improves the segmentation of hepatic vascular networks in ultrasound volumes using a three-dimensional U-Net with intensity transformation-based data augmentation, a strategy that artificially diversifies training images by altering their brightness and contrast distributions, thereby teaching the network to be robust to the notoriously variable gain settings of ultrasound machines. Two further liver studies move from anatomy to diagnosis. One performs simultaneous grading of liver fibrosis, inflammation, and steatosis using multimodal quantitative ultrasound within an artificial intelligence framework. Quantitative ultrasound extracts numerical features, such as acoustic attenuation and backscatter statistics, from the raw image signal rather than from its visual appearance, and combining these features across modalities allows a single examination to characterize several dimensions of liver disease at once. This matters clinically because fatty liver disease and its inflammatory and fibrotic progression are among the most common chronic conditions worldwide, and current assessment often relies on liver biopsy, an invasive procedure that samples only a tiny fraction of the organ. Another liver paper targets nonalcoholic steatohepatitis specifically, using convolutional neural networks with a dual-branch architecture that fuses global and local image features, an appropriate design for a disease in which both diffuse texture changes and focal patterns carry diagnostic meaning.

Vascular structure segmentation receives dedicated attention through a network that encodes A-line coordinates, the one-dimensional signal traces that underlie ultrasound beam formation, for the purpose of delineating vascular anatomy. Working at the level of A-lines rather than finished images represents a return to the physical roots of ultrasound, and it suggests that some of the field’s future gains may come from integrating signal-level information with image-level learning. In a related application, deep learning-based segmentation of abdominal aortic aneurysms and their intraluminal thrombus in three-dimensional ultrasound images addresses a condition whose surveillance currently depends on repeated measurements of a dilating vessel. Automated and reproducible measurement of both the aneurysm sac and the blood clot within it could standardize the growth assessments that determine when surgical intervention becomes necessary.

Oncological imaging appears in several guises. Breast ultrasound, a mainstay of screening in dense tissue where mammography performs poorly, is the subject of two contributions: one combines adaptive ensemble loss functions with multi-scale attention in a U-shaped segmentation network, and another applies hybrid adaptive attention with deep supervision to breast lesion segmentation in ultrasound computed tomography images, a modality that reconstructs acoustic properties of tissue in three dimensions. A separate study asks a deceptively simple question: can incorporating image resolution into neural networks improve kidney tumor classification performance in ultrasound images? The affirmative answer carries a broader lesson, namely that metadata and acquisition parameters that are usually discarded during preprocessing can themselves be informative features, encoding information about tissue scattering physics that the image alone may not fully convey.

The collection also reaches into specialties where ultrasound plays a surgical rather than diagnostic role. One paper presents three-dimensional ultrasound shape completion and anatomical feature detection for minimally invasive spine surgery. In such procedures, the surgeon works through small incisions with limited direct visualization, and intraoperative ultrasound offers a radiation-free alternative to fluoroscopy for navigating the bony anatomy. Shape completion, the task of inferring the full geometry of a structure from partial or occluded observations, is a natural fit for this setting because acoustic shadowing from bone routinely leaves gaps in the image. Another contribution introduces a network for tongue contour segmentation built on a carefully managed dataset, extending ultrasound into the assessment of speech and swallowing, where tracking the tongue’s shape during movement has applications ranging from speech therapy to the study of deglutition disorders.

Taken together, these twenty studies sketch the contours of a field in transition. Several unifying threads deserve emphasis. First, foundation models and promptable segmentation architectures, exemplified by the adaptation of the Segment Anything Model to cardiac imaging, are lowering the cost of building application-specific tools, since a pretrained general model can be specialized with far less data than a network trained from scratch. Second, semi-supervised and data-efficient paradigms are ascendant precisely because expert annotation remains the scarcest resource in medical imaging; every hour a cardiologist spends tracing ventricles is an hour not spent with patients. Third, uncertainty quantification, whether through evidential training or related techniques, is moving from a theoretical nicety to a practical requirement, since a clinical assistant that cannot signal when it is unsure is more dangerous than one that abstains. Fourth, the papers collectively span the pipeline from raw signal to deployed system, encompassing data augmentation, architecture design, loss engineering, and operational maintenance, which reflects a maturing understanding that clinical impact requires excellence at every stage rather than at any single one.

The geographic and institutional diversity of the authorship, spanning universities across Italy and an international community of contributors, underscores that ultrasound artificial intelligence is a genuinely global enterprise. Ultrasound is the imaging modality most accessible in low-resource settings, and advances that make its interpretation more reliable without requiring expensive hardware have direct implications for health equity. A portable probe paired with a smartphone and a well-calibrated neural network can, in principle, bring diagnostic capabilities to places where computed tomography scanners and specialist radiologists are unavailable. The editorial’s framing of the special issue as a revolution in image processing and analysis is therefore not merely rhetorical: it points toward a future in which the operator dependence that has long been ultrasound’s greatest limitation is progressively offset by algorithms that standardize measurement, flag pathology, and guide the hand of the person holding the probe. The papers gathered under the AI4US banner, now available as an online collection, provide both a snapshot of current capability and a roadmap of the problems, from four-dimensional valve dynamics to quantitative tissue characterization, that the next generation of work will need to solve.

Subject of Research: The application of artificial intelligence and deep learning to ultrasound image processing and analysis across medical specialties.

Article Title: AI4US: Revolutionizing ultrasound image processing and analysis with artificial intelligence

Article References: Moccia, S., Fiorentino, M. C., & Tomassini, S. (2026). AI4US: Revolutionizing ultrasound image processing and analysis with artificial intelligence. Medical & Biological Engineering & Computing. https://doi.org/10.1007/s11517-026-03667-0

Image Credits: AI Generated

DOI: 10.1007/s11517-026-03667-0

Keywords: ultrasound, artificial intelligence, deep learning, echocardiography, image segmentation, medical imaging, machine learning, computer vision, fetal ultrasound, breast ultrasound, SAM, clinical diagnosis