Illuminating the future of precision cancer surgery with fluorescence imaging and artificial intelligence convergence

npj Precision Oncology, Journal Year: 2024, Volume and Issue: 8(1)

Published: Sept. 9, 2024

Language: Английский

---

Artificial Intelligence for context-aware surgical guidance in complex robot-assisted oncological procedures: An exploratory feasibility study

European Journal of Surgical Oncology, Journal Year: 2023, Volume and Issue: 50(12), P. 106996 - 106996

Published: July 28, 2023

Complex oncological procedures pose various surgical challenges including dissection in distinct tissue planes and preservation of vulnerable anatomical structures throughout different phases. In rectal surgery, violation increases the risk local recurrence autonomous nerve damage resulting incontinence sexual dysfunction. This work explores feasibility phase recognition target structure segmentation robot-assisted resection (RARR) using machine learning.A total 57 RARR were recorded subsets these annotated with respect to phases exact locations (anatomical structures, types, static areas). For recognition, three learning models trained: LSTM, MSTCN, Trans-SVNet. Based on pixel-wise annotations 9037 images, individual based DeepLabv3 trained. Model performance was evaluated F1 score, Intersection-over-Union (IoU), accuracy, precision, recall, specificity.The best results for achieved MSTCN model (F1 score: 0.82 ± 0.01, accuracy: 0.84 0.03). Mean IoUs ranged from 0.14 0.22 0.80 organs types 0.11 0.44 0.30 areas. Image quality, distorting factors (i.e. blood, smoke), technical lack depth perception) considerably impacted performance.Machine learning-based selected are feasible RARR. future, such functionalities could be integrated into a context-aware guidance system surgery.

Language: Английский

---

Self-supervised Learning for Endoscopic Video Analysis

Lecture notes in computer science, Journal Year: 2023, Volume and Issue: unknown, P. 569 - 578

Published: Jan. 1, 2023

Language: Английский

---

Deep learning in surgical process Modeling: A systematic review of workflow recognition

Journal of Biomedical Informatics, Journal Year: 2025, Volume and Issue: unknown, P. 104779 - 104779

Published: Jan. 1, 2025

Language: Английский

---

A parameter-free self-training algorithm based on the three successive confirmation rule

Engineering Applications of Artificial Intelligence, Journal Year: 2025, Volume and Issue: 144, P. 110165 - 110165

Published: Feb. 1, 2025

Language: Английский

---

Artificial intelligence-assisted phase recognition and skill assessment in laparoscopic surgery: a systematic review

Frontiers in Surgery, Journal Year: 2025, Volume and Issue: 12

Published: April 11, 2025

With the widespread adoption of minimally invasive surgery, laparoscopic surgery has been an essential component modern surgical procedures. As key technologies, phase recognition and skill evaluation aim to identify different stages process assess surgeons’ operational skills using automated methods. This, in turn, can improve quality surgeons. This review summarizes progress research recognition, evaluation. At first, importance is introduced, clarifying relationship between evaluation, other tasks. The publicly available datasets for tasks are then detailed. highlights methods that have exhibited superior performance these public identifies common characteristics high-performing Based on insights obtained, commonly used models this field summarized. In addition, study briefly outlines standards evaluating skills. Finally, analysis difficulties researchers face potential future development directions presented. Moreover, paper aims provide valuable references researchers, promoting further advancements domain.

Language: Английский

---

Vision techniques for anatomical structures in laparoscopic surgery: a comprehensive review

Frontiers in Surgery, Journal Year: 2025, Volume and Issue: 12

Published: April 14, 2025

Laparoscopic surgery is the method of choice for numerous surgical procedures, while it confronts a lot challenges. Computer vision exerts vital role in addressing these challenges and has become research hotspot, especially classification, segmentation, target detection abdominal anatomical structures. This study presents comprehensive review last decade this area. At first, categorized overview core subtasks presented regarding their relevance applicability to real-world medical scenarios. Second, dataset used experimental validation statistically analyzed. Subsequently, technical approaches trends tasks are explored detail, highlighting advantages, limitations, practical implications. Additionally, evaluation methods three types discussed. Finally, gaps current identified. Meanwhile, great potential development area emphasized.

Language: Английский

---

Early operative difficulty assessment in laparoscopic cholecystectomy via snapshot-centric video analysis

International Journal of Computer Assisted Radiology and Surgery, Journal Year: 2025, Volume and Issue: unknown

Published: April 21, 2025

Language: Английский

---

Digital twins as a unifying framework for surgical data science: the enabling role of geometric scene understanding

Artificial Intelligence Surgery, Journal Year: 2024, Volume and Issue: 4(3), P. 109 - 38

Published: July 5, 2024

Surgical data science is devoted to enhancing the quality, safety, and efficacy of interventional healthcare. While use powerful machine learning algorithms becoming standard approach for surgical science, underlying end-to-end task models directly infer high-level concepts (e.g., phase or skill) from low-level observations endoscopic video). This nature contemporary approaches makes vulnerable non-causal relationships in requires re-development all components if new tasks are be solved. The digital twin (DT) paradigm, an building maintaining computational representations real-world scenarios, offers a framework separating processing inference. In DT paradigm would allow development generalist on top universal representation, deferring model computer vision algorithms. this latter effort creation, geometric scene understanding plays central role updating model. work, we visit existing representations, tasks, successful applications primitive frameworks. Although advanced methods still hindered by lack annotations, complexity limited observability scene, emerging works synthetic generation, sim-to-real generalization, foundation offer directions overcoming these challenges advancing paradigm.

Language: Английский

---

Foundation models in gastrointestinal endoscopic AI: Impact of architecture, pre-training approach and data efficiency

Medical Image Analysis, Journal Year: 2024, Volume and Issue: 98, P. 103298 - 103298

Published: Aug. 12, 2024

Pre-training deep learning models with large data sets of natural images, such as ImageNet, has become the standard for endoscopic image analysis. This approach is generally superior to training from scratch, due scarcity high-quality medical imagery and labels. However, it still unknown whether learned features on provide an optimal starting point downstream imaging tasks. Intuitively, pre-training closer target domain could lead better-suited feature representations. study evaluates leveraging in-domain in gastrointestinal analysis potential benefits compared images. To this end, we present a dataset comprising 5,014,174 images eight different centers (GastroNet-5M), exploit self-supervised SimCLRv2, MoCov2 DINO learn relevant The are derived multiple methods, variable amounts and/or labels (e.g. Billion-scale semi-weakly supervised ImageNet-21k). effects evaluation performed five sets, particularly designed variety tasks, example, GIANA angiodyplsia detection Kvasir-SEG polyp segmentation. findings indicate that domain-specific pre-training, specifically using framework, results into better performing any On ResNet50 Vision-Transformer-small architectures, utilizing leads average performance boost 1.63% 4.62%, respectively, datasets. improvement measured against best achieved through within evaluated frameworks. Moreover, pre-trained also exhibit increased robustness distortion perturbations (noise, contrast, blur, etc.), where 1.28% 3.55% higher metrics, found Overall, highlights importance improving generic nature, scalability GastroNet-5M weights made publicly available our repository: huggingface.co/tgwboers/GastroNet-5M_Pretrained_Weights.

Language: Английский

---