Reporting of colorectal polyp morphology using the Paris classification is often inaccurate. Multimodal large language models (M-LLMs) may support morphological assessment. This study aimed to evaluate the accuracy of an M-LLM (GPT-4o) in classifying colorectal polyp morphology compared with expert and non-expert endoscopists.We used the SUN dataset of colonoscopy videos from 100 unique colorectal polyps, each labeled with the validated Paris classification. An M-LLM (GPT-4o) classified five representative frames per lesion. Three expert and three non-expert endoscopists, blinded to one another, performed the same task. The primary outcome was accuracy in differentiating non-polypoid (IIa/IIc) from polypoid (Is/Ip/Isp) lesions. The secondary outcome was accuracy in differentiating sessile (Is) from pedunculated (Ip/Isp) lesions. Given the exploratory design, no multiplicity correction was applied; point estimates are presented with 95% confidence intervals (CIs), and P values are interpreted descriptively.M-LLM accuracy for differentiating non-polypoid from polypoid lesions was 73% (95% CI 63%-81%), comparable to experts (75%, 65%-83%; P = 0.84) and non-experts (77%, 68%-85%; P = 0.52), with similar sensitivity and specificity. Accuracy for differentiating sessile from pedunculated lesions was 55% (95% CI 42%-67%), lower than experts (76%; P = 0.02) and non-experts (77%; P = 0.01), primarily due to poor specificity (12% vs. experts 82% and non-experts 88%; P < 0.01 for both comparisons).M-LLMs performed comparably to endoscopists in distinguishing non-polypoid from polypoid lesions but failed to reliably identify pedunculated morphology.
21 January 2026 in Computer Methods and Programs in Biomedicine
Accurate assessment of colorectal lesion morphology during colonoscopy is essential for guiding treatment and estimating cancer risk. The Paris classification is widely adopted for this purpose but suffers from substantial inter-observer variability, while Vision Transformers (ViTs) can base their decisions on diffuse, off-lesion attention patterns that are hard to interpret. This study investigates whether directly supervising ViT attention maps with expert lesion annotations can concurrently improve Paris classification performance and model explainability. We propose a Lesion-Focused Attention Loss (LLFA), an attention-supervised pretraining objective that uses expert polyp bounding boxes to focus last-layer [CLS] attention on annotated lesion regions, followed by standard cross-entropy fine-tuning. LLFA is applied to six ViT architectures and evaluated on the public SUN dataset for binary (0-I vs. 0-II) and three-class (0-Ip, 0-Is, 0-IIa) Paris classification. Performance is assessed using frame-wise accuracy and the AttIn, we additionally perform an ablation study against a Grad-CAM consistency baseline. Attention-supervised pretraining yields consistent gains in both accuracy and lesion-focused attention. Across the six ViTs, adding LLFA improves three-class accuracy by up to 7 percentage points. In a detailed ablation on ViT-B/16, LLFA outperforms a Grad-CAM consistency baseline by about 5-13 percentage points across the 2-class and 3-class tasks, and χ2 tests confirm a significant association between high AttIn and correct predictions. Direct supervision of ViT attention with LLFA leverages expert knowledge to jointly boost Paris classification accuracy and spatial interpretability, and compares favourably with Grad-CAM-based explanation regularisation. The source code and dataset splits are publicly available at https://github.com/LucaCarlini/SUNDatasetPretraining.
Interface tissues, such as the enthesis connecting ligaments to bone, present multiphasic architectures with continuous gradients in structure, composition, and mechanics. Engineering such complex transitions remains a major challenge in biofabrication. This study aims to develop a hybrid manufacturing and machine learning (ML)-guided design strategy to create functionally graded scaffolds for anterior cruciate ligament (ACL) reconstruction. A hybrid biofabrication platform was used to integrate extrusion-based three-dimensional printing and electrospinning within a single workflow. Polycaprolactone was used as the common biomaterial for both modalities. Four scaffold designs, varying in electrospun midsection length, slit patterning, and core geometry, were fabricated to replicate the native ACL’s zonal architecture. Scaffolds were characterized through scanning electron microscopy (SEM) and uniaxial tensile testing. Resulting data were used to train a ML model to predict mechanical performance from geometric features. The model was then used to generate a fifth scaffold design optimized for enhanced performance. The hybrid process successfully fabricated multiscale scaffolds with integrated bone-like, enthesis-like, and ligament-like regions. SEM confirmed morphological integration between printed and electrospun structures. Mechanical testing revealed design-dependent variations in strength and stiffness. The ML model identified slit number and outer diameter as key predictors and guided the design of an optimized scaffold that combined the compliance of slitted geometries with enhanced mechanical strength. The ML-optimized scaffold achieved the highest tensile force among the slitted designs and improved stiffness compared to the other slitted configurations this study demonstrates a predictive and performance-driven biofabrication strategy that integrates hybrid additive manufacturing and ML. The approach enables rational scaffold optimization, reduces empirical iterations, and supports the development of biomimetic constructs for soft-to-hard tissue engineering. While focused on ACL reconstruction, the workflow is adaptable to a wide range of tissue interfaces.
Background/Objective: High-frequency micro-ultrasound (micro-US) offers real-time, high-resolution imaging for prostate cancer. Although artificial intelligence (AI) has shown potential in enhancing micro-US interpretation, a comprehensive review of this emerging field is currently missing. This review synthesizes current evidence on AI applied to ExactVu 29 MHz micro-US for prostate cancer. Methods: PubMed/MEDLINE, Embase, Scopus, Web of Science and the Cochrane Library were searched up to December 2025. Studies were included if they applied machine learning or deep learning directly to 29 MHz micro-US data and reported quantitative performance metrics. Results: Ten studies met the inclusion criteria: six on prostate cancer detection, three on prostate segmentation and one on micro-US–histopathology registration. Detection models ranged from classical quantitative ultrasound machine learning to deep architectures using self-supervision, transformers, multiple-instance learning, ensemble calibration and 3D segmentation-based pipelines. Among core-level models for clinically significant cancer, area under the receiver operating characteristic curve (AUROC) values clustered around 0.76–0.81; one lesion-level framework reported an AUROC of 0.92, though at a non-comparable analytical unit. Segmentation studies achieved accurate prostate delineation (Dice similarity coefficient ≈ 0.94), and a single study demonstrated high-precision 3D registration to whole-mount histopathology (Dice similarity coefficient 0.97 and landmark error < 3 mm). All studies evaluated AI on previously acquired data, without real-time clinical implementation. Conclusions: AI for micro-US shows promising and reproducible early results across detection, segmentation and registration, but evidence is still limited. In view of the potential of AI to optimize micro-US utilization and its related advantages, additional efforts are warranted to achieve clinical adoption.