Segment anything model for medical images?

Medical Image Analysis, Год журнала: 2023, Номер 92, С. 103061 - 103061

Опубликована: Дек. 7, 2023

Язык: Английский

---

SegRap2023: A benchmark of organs-at-risk and gross tumor volume Segmentation for Radiotherapy Planning of Nasopharyngeal Carcinoma

Medical Image Analysis, Год журнала: 2025, Номер 101, С. 103447 - 103447

Опубликована: Янв. 2, 2025

Язык: Английский

---

Towards more precise automatic analysis: a systematic review of deep learning-based multi-organ segmentation

BioMedical Engineering OnLine, Год журнала: 2024, Номер 23(1)

Опубликована: Июнь 8, 2024

Abstract Accurate segmentation of multiple organs in the head, neck, chest, and abdomen from medical images is an essential step computer-aided diagnosis, surgical navigation, radiation therapy. In past few years, with a data-driven feature extraction approach end-to-end training, automatic deep learning-based multi-organ methods have far outperformed traditional become new research topic. This review systematically summarizes latest this field. We searched Google Scholar for papers published January 1, 2016 to December 31, 2023, using keywords “multi-organ segmentation” “deep learning”, resulting 327 papers. followed PRISMA guidelines paper selection, 195 studies were deemed be within scope review. summarized two main aspects involved segmentation: datasets methods. Regarding datasets, we provided overview existing public conducted in-depth analysis. Concerning methods, categorized approaches into three major classes: fully supervised, weakly supervised semi-supervised, based on whether they require complete label information. achievements these terms accuracy. discussion conclusion section, outlined current trends segmentation.

Язык: Английский

---

HaN-Seg: The head and neck organ-at-risk CT and MR segmentation challenge

Radiotherapy and Oncology, Год журнала: 2024, Номер 198, С. 110410 - 110410

Опубликована: Июнь 24, 2024

To promote the development of auto-segmentation methods for head and neck (HaN) radiation treatment (RT) planning that exploit information computed tomography (CT) magnetic resonance (MR) imaging modalities, we organized HaN-Seg: The Head Neck Organ-at-Risk CT MR Segmentation Challenge. challenge task was to automatically segment 30 organs-at-risk (OARs) HaN region in 14 withheld test cases given availability 42 publicly available training cases. Each case consisted one contrast-enhanced T1-weighted image same patient, with up corresponding reference OAR delineation masks. performance evaluated terms Dice similarity coefficient (DSC) 95-percentile Hausdorff distance (HD95), statistical ranking applied each metric by pairwise comparison submitted using Wilcoxon signed-rank test. While 23 teams registered challenge, only seven their final phase. top-performing team achieved a DSC 76.9 % HD95 3.5 mm. All participating utilized architectures based on U-Net, winning leveraging rigid registration combined network entry-level concatenation both modalities. This simulated real-world clinical scenario providing non-registered images varying fields-of-view voxel sizes. Remarkably, segmentation surpassing inter-observer agreement dataset. These results set benchmark future research this dataset paired multi-modal general.

Язык: Английский

---

vOARiability: Interobserver and intermodality variability analysis in OAR contouring from head and neck CT and MR images

Medical Physics, Год журнала: 2024, Номер 51(3), С. 2175 - 2186

Опубликована: Янв. 17, 2024

Abstract Background Accurate and consistent contouring of organs‐at‐risk (OARs) from medical images is a key step radiotherapy (RT) cancer treatment planning. Most approaches rely on computed tomography (CT) images, but the integration complementary magnetic resonance (MR) modality highly recommended, especially perspective OAR contouring, synthetic CT MR image generation for MR‐only RT, MR‐guided RT. Although has been recognized as valuable OARs in head neck (HaN) region, accuracy consistency resulting contours have not yet objectively evaluated. Purpose To analyze interobserver intermodality variability HaN performed by observers with different level experience same patients. Methods In final cohort 27 patients, up to 31 were obtained radiation oncology resident (junior observer, JO) board‐certified oncologist (senior SO). The then evaluated terms variability, characterized agreement among (JO SO) when selected (CT or MR), modalities MR) contoured observer SO), both Dice coefficient (DC) 95‐percentile Hausdorff distance (HD ). Results mean (±standard deviation) was 69.0 ± 20.2% 5.1 4.1 mm, while 61.6 19.0% 6.1 4.3 mm DC HD , respectively, across all OARs. Statistically significant differences only found specific registration resulted target error 1.7 0.5 which considered valid analysis variability. Conclusions was, general, similar modalities, did considerably affect performance. However, results indicate that an difficult contour regardless whether it image, may be important factor are deemed contour. Several can also attributed adherence guidelines, poor visibility without distinctive boundaries either images. considerable observed OARs, concluded almost degree modality, works favor

Язык: Английский

---

ScribblePrompt: Fast and Flexible Interactive Segmentation for Any Biomedical Image

Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 207 - 229

Опубликована: Ноя. 9, 2024

Язык: Английский

---

Essential parameters needed for a U-Net-based segmentation of individual bones on planning CT images in the head & neck region using limited datasets for radiotherapy application

Physics in Medicine and Biology, Год журнала: 2024, Номер 69(3), С. 035008 - 035008

Опубликована: Янв. 2, 2024

The field of radiotherapy is highly marked by the lack datasets even with availability public datasets. Our study uses a very limited dataset to provide insights on essential parameters needed automatically and accurately segment individual bones planning CT images head neck cancer patients.

Язык: Английский

---

Segmentation of 71 Anatomical Structures Necessary for the Evaluation of Guideline-Conforming Clinical Target Volumes in Head and Neck Cancers

Cancers, Год журнала: 2024, Номер 16(2), С. 415 - 415

Опубликована: Янв. 18, 2024

The delineation of the clinical target volumes (CTVs) for radiation therapy is time-consuming, requires intensive training and shows high inter-observer variability. Supervised deep-learning methods depend heavily on consistent data; thus, State-of-the-Art research focuses making CTV labels more homogeneous strictly bounding them to current standards. International consensus expert guidelines standardize by conditioning extension volume surrounding anatomical structures. Training strategies that directly follow construction rules given in or possibility quantifying conformance manually drawn contours are still missing. Seventy-one structures relevant head- neck-cancer patients, according guidelines, were segmented 104 computed tomography scans, assess automating their segmentation deep learning methods. All 71 subdivided into three subsets non-overlapping structures, a 3D nnU-Net model with five-fold cross-validation was trained each subset, automatically segment planning scans. We report DICE, Hausdorff distance surface DICE + 5 most which no previous accuracies have been reported. For those prediction values reported, our accuracy matched exceeded reported values. predictions from models always better than predicted TotalSegmentator. sDICE 2 mm margin larger 80% almost all Individual decreased analyzed discussed respect impact following guidelines. No deviation expected affect rule-based automation delineation.

Язык: Английский

---

Transfer learning for auto‐segmentation of 17 organs‐at‐risk in the head and neck: Bridging the gap between institutional and public datasets

Medical Physics, Год журнала: 2024, Номер 51(7), С. 4767 - 4777

Опубликована: Фев. 20, 2024

Abstract Background Auto‐segmentation of organs‐at‐risk (OARs) in the head and neck (HN) on computed tomography (CT) images is a time‐consuming component radiation therapy pipeline that suffers from inter‐observer variability. Deep learning (DL) has shown state‐of‐the‐art results CT auto‐segmentation, with larger more diverse datasets showing better segmentation performance. Institutional auto‐segmentation have been small historically (n < 50) due to time required for manual curation anatomical labels. Recently, large public > 1000 aggregated) become available through online repositories such as The Cancer Imaging Archive. Transfer technique applied when training samples are scarce, but dataset closely related domain available. Purpose purpose this study was investigate whether could be used place an institutional 500), or augment performance via transfer learning, building HN OAR models use. Methods were trained (public models) smaller (institutional models). fine‐tuned using (transfer We assessed both model generalizability by comparison models. Additionally, effect size investigated. All DL high‐resolution, two‐stage architecture based popular 3D U‐Net. Model evaluated five geometric measures: dice similarity coefficient (DSC), surface DSC, 95 th percentile Hausdorff distance, mean distance (MSD), added path length. Results For subset OARs (left/right optic nerve, spinal cord, left submandibular), performed significantly ( p 0.05) than, showed no significant difference to, under most metrics examined. remaining OARs, inferior models, although differences (DSC ≤ 0.03, MSD 0.5 mm) seven (brainstem, left/right lens, parotid, mandible, right submandibular). than cord) margin improvement 0.02, 0.4 mm). When numbers limited, outperformed Conclusion Training data alone suitable number OARs. Using only incurred deficit other compared alone, may preferable over dataset. available, pretrained provided modest several model, beneficial

Язык: Английский

---

Towards Automation in Radiotherapy Planning: A Deep Learning Approach for the Delineation of Parotid Glands in Head and Neck Cancer

Bioengineering, Год журнала: 2024, Номер 11(3), С. 214 - 214

Опубликована: Фев. 24, 2024

The delineation of parotid glands in head and neck (HN) carcinoma is critical to assess radiotherapy (RT) planning. Segmentation processes ensure precise target position treatment precision, facilitate monitoring anatomical changes, enable plan adaptation, enhance overall patient safety. In this context, artificial intelligence (AI) deep learning (DL) have proven exceedingly effective precisely outlining tumor tissues and, by extension, the organs at risk. This paper introduces a DL framework using AttentionUNet neural network for automatic gland segmentation HN cancer. Extensive evaluation model performed two public one private dataset, while accuracy compared with other state-of-the-art schemas. To replanning necessity during treatment, an additional registration method implemented on output, aligning images different modalities (Computed Tomography (CT) Cone Beam CT (CBCT)). outperforms similar methods (Dice Similarity Coefficient: 82.65% ± 1.03, Hausdorff Distance: 6.24 mm 2.47), confirming its effectiveness. Moreover, subsequent procedure displays increased similarity, providing insights into effects RT procedures planning adaptations. implementation proposed indicates effectiveness not only structures, but also provision information adaptive support.

Язык: Английский

---