Development and routine implementation of deep learning algorithm for automatic brain metastases segmentation on MRI for RANO-BM criteria follow-up DOI Creative Commons

Loïse Dessoude,

Raphaëlle Lemaire,

Roger Y. Andres

et al.

NeuroImage, Journal Year: 2025, Volume and Issue: 306, P. 121002 - 121002

Published: Jan. 10, 2025

The RANO-BM criteria, which employ a one-dimensional measurement of the largest diameter, are imperfect due to fact that lesion volume is neither isotropic nor homogeneous. Furthermore, this approach inherently time-consuming. Consequently, in clinical practice, monitoring patients trials compliance with criteria rarely achieved. objective study was develop and validate an AI solution capable delineating brain metastases (BM) on MRI easily obtain, using in-house solution, as well BM routine setting. A total 27,456 post-Gadolinium-T1 from 132 were employed study. deep learning (DL) model constructed PyTorch Lightning frameworks, UNETR transfer method segment MRI. visual analysis results demonstrates confident delineation lesions. shows 100 % accuracy predicting comparison expert medical doctor. There high degree overlap between doctor's segmentation, mean DICE score 0.77. diameter lesions found be concordant reference segmentation. user interface developed can readily provide following accessible everyone without expertise offers effective segmentation substantial time savings.

Language: Английский

A novel lightweight deep convolutional neural network for early detection of oral cancer DOI
Fahed Jubair,

Omar Al‐karadsheh,

Dimitrios Malamos

et al.

Oral Diseases, Journal Year: 2021, Volume and Issue: 28(4), P. 1123 - 1130

Published: Feb. 26, 2021

To develop a lightweight deep convolutional neural network (CNN) for binary classification of oral lesions into benign and malignant or potentially using standard real-time clinical images.A small CNN, that uses pretrained EfficientNet-B0 as transfer learning model, was proposed. A data set 716 images used to train test the proposed model. Accuracy, specificity, sensitivity, receiver operating characteristics (ROC) area under curve (AUC) were evaluate performance. Bootstrapping with 120 repetitions calculate arithmetic means 95% confidence intervals (CIs).The CNN model achieved an accuracy 85.0% (95% CI: 81.0%-90.0%), specificity 84.5% 78.9%-91.5%), sensitivity 86.7% 80.4%-93.3%) AUC 0.928 0.88-0.96).Deep CNNs can be effective method build low-budget embedded vision devices limited computation power memory capacity diagnosis cancer. Artificial intelligence (AI) improve quality reach cancer screening early detection.

Language: Английский

Citations

121
An artificial intelligence framework and its bias for brain tumor segmentation: A narrative review DOI
Suchismita Das, Gopal Krishna Nayak, Luca Saba

et al.

Computers in Biology and Medicine, Journal Year: 2022, Volume and Issue: 143, P. 105273 - 105273

Published: Feb. 19, 2022

Language: Английский

Citations

107
Predicting EGFR Mutation Status in Non–Small Cell Lung Cancer Using Artificial Intelligence: A Systematic Review and Meta-Analysis DOI
Hung Song Nguyen, Dang Khanh Ngan Ho, Nam Nhat Nguyen

et al.

Academic Radiology, Journal Year: 2023, Volume and Issue: 31(2), P. 660 - 683

Published: April 28, 2023

Language: Английский

Citations

60
Quality assessment standards in artificial intelligence diagnostic accuracy systematic reviews: a meta-research study DOI Creative Commons
Shruti Jayakumar, Viknesh Sounderajah, Pasha Normahani

et al.

npj Digital Medicine, Journal Year: 2022, Volume and Issue: 5(1)

Published: Jan. 27, 2022

Artificial intelligence (AI) centred diagnostic systems are increasingly recognised as robust solutions in healthcare delivery pathways. In turn, there has been a concurrent rise secondary research studies regarding these technologies order to influence key clinical and policymaking decisions. It is therefore essential that accurately appraise methodological quality risk of bias within shortlisted trials reports. assess whether this critical step performed, we undertook meta-research study evaluating adherence the Quality Assessment Diagnostic Accuracy Studies 2 (QUADAS-2) tool AI accuracy systematic reviews. A literature search was conducted on all published from 2000 December 2020. Of 50 included reviews, 36 performed assessment, which 27 utilised QUADAS-2 tool. Bias reported across four domains QUADAS-2. Two hundred forty-three 423 (57.5%) reviews utilising high or unclear patient selection domain, 110 (26%) index test 121 (28.6%) reference standard domain 157 (37.1%) flow timing domain. This demonstrates incomplete uptake assessment tools AI-based highlights inconsistent reporting assessment. Poor standards act barriers implementation. The creation an AI-specific extension for may facilitate safe translation into practice.

Language: Английский

Citations

67
Small-cell lung cancer brain metastasis: From molecular mechanisms to diagnosis and treatment DOI Creative Commons
Yingze Zhu,

Yishuang Cui,

Xuan Zheng

et al.

Biochimica et Biophysica Acta (BBA) - Molecular Basis of Disease, Journal Year: 2022, Volume and Issue: 1868(12), P. 166557 - 166557

Published: Sept. 24, 2022

Lung cancer is the most malignant human worldwide, also with highest incidence rate. However, small-cell lung (SCLC) accounts for 14 % of all cases. Approximately 10 patients SCLC have brain metastasis at time diagnosis, which leading cause death worldwide. The median overall survival only 4.9 months, and a long-tern cure exists due to limited common therapeutic options. Recent studies enhanced our understanding molecular mechanisms meningeal metastasis, multimodality treatments brought new hopes better disease. This review aimed offer an insight into cellular processes different metastatic stages revealed by established animal models, major diagnostic methods SCLC. Additionally, it provided in-depth information on recent advances in treatments, highlighted several models biomarkers promises improve prognosis

Language: Английский

Citations

48
A comprehensive dataset of annotated brain metastasis MR images with clinical and radiomic data DOI Creative Commons
Beatriz Ocaña-Tienda, Julián Pérez-Beteta,

José Villanueva-García

et al.

Scientific Data, Journal Year: 2023, Volume and Issue: 10(1)

Published: April 14, 2023

Brain metastasis (BM) is one of the main complications many cancers, and most frequent malignancy central nervous system. Imaging studies BMs are routinely used for diagnosis disease, treatment planning follow-up. Artificial Intelligence (AI) has great potential to provide automated tools assist in management disease. However, AI methods require large datasets training validation, date there have been just publicly available imaging dataset 156 BMs. This paper publishes 637 high-resolution 75 patients harboring 260 BM lesions, their respective clinical data. It also includes semi-automatic segmentations 593 BMs, including pre- post-treatment T1-weighted cases, a set morphological radiomic features cases segmented. data-sharing initiative expected enable research into performance evaluation automatic detection, lesion segmentation, disease status as well development validation predictive prognostic with applicability.

Language: Английский

Citations

30
Artificial Intelligence in Brain Tumour Surgery—An Emerging Paradigm DOI Open Access
Simon C. Williams, Hugo Layard Horsfall, Jonathan P. Funnell

et al.

Cancers, Journal Year: 2021, Volume and Issue: 13(19), P. 5010 - 5010

Published: Oct. 7, 2021

Artificial intelligence (AI) platforms have the potential to cause a paradigm shift in brain tumour surgery. Brain surgery augmented with AI can result safer and more effective treatment. In this review article, we explore current future role of patients undergoing surgery, including aiding diagnosis, optimising surgical plan, providing support during operation, better predicting prognosis. Finally, discuss barriers successful clinical implementation, ethical concerns, provide our perspective on how field could be advanced.

Language: Английский

Citations

53
Development and validation of a deep-learning model for detecting brain metastases on 3D post-contrast MRI: a multi-center multi-reader evaluation study DOI Creative Commons

Shaohan Yin,

Xiao Luo,

Yadi Yang

et al.

Neuro-Oncology, Journal Year: 2022, Volume and Issue: 24(9), P. 1559 - 1570

Published: Jan. 27, 2022

Accurate detection is essential for brain metastasis (BM) management, but manual identification laborious. This study developed, validated, and evaluated a BM (BMD) system.Five hundred seventy-three consecutive patients (10 448 lesions) with newly diagnosed BMs 377 without were retrospectively enrolled to develop multi-scale cascaded convolutional network using 3D-enhanced T1-weighted MR images. BMD was validated prospective validation set comprising an internal (46 349 lesions; 44 BMs) three external sets (102 717 108 BMs). The lesion-based sensitivity the number of false positives (FPs) per patient analyzed. reading time trainees experienced radiologists from hospitals set.The FPs 95.8% 0.39 in test set, 96.0% 0.27 ranged 88.9% 95.5% 0.29 0.66 sets. system achieved higher (93.2% [95% CI, 91.6-94.7%]) than all (ranging 68.5% 65.7-71.3%] 80.4% 78.0-82.8%], P < .001). Radiologist improved BMD, reaching 92.7% 95.0%. mean reduced by 47% 32% assisted relative that BMD.BMD enables accurate detection. Reading improves radiologists' reduces their times.

Language: Английский

Citations

32
Deep learning for brain metastasis detection and segmentation in longitudinal MRI data DOI
Yixing Huang, Christoph Bert, Philipp Sommer

et al.

Medical Physics, Journal Year: 2022, Volume and Issue: 49(9), P. 5773 - 5786

Published: July 14, 2022

Brain metastases (BM) occur frequently in patients with metastatic cancer. Early and accurate detection of BM is essential for treatment planning prognosis radiation therapy. Due to their tiny sizes relatively low contrast, small are very difficult detect manually. With the recent development deep learning technologies, several res earchers have reported promising results automated brain metastasis detection. However, sensitivity still not high enough BM, integration into clinical practice regard differentiating true from false positives (FPs) challenging.The DeepMedic network binary cross-entropy (BCE) loss used as our baseline method. To improve performance, a custom called volume-level sensitivity-specificity (VSS) proposed, which rates specificity at (sub)volume level. As precision always trade-off, either or can be achieved by adjusting weights VSS without decline dice score coefficient segmented metastases. reduce metastasis-like structures being detected FP metastases, temporal prior volume proposed an additional input DeepMedic. The modified DeepMedic+ distinction. Combining high-sensitivity DeepMedic+, majority positive confirmed specificity, while candidates each patient marked detailed expert evaluation.Our improves detection, increasing 85.3% BCE 97.5% VSS. Alternatively, improved 69.1% 98.7% Comparing same loss, 44.4% reduced model reaches 99.6% high-specificity model. mean all about 0.81. ensemble models, on average only 1.5 per need further check, confirmed.Our precision. able distinguish confidence that require special review follow-up, particularly well-fit requirements support real practice. This facilitates segmentation neuroradiologists diagnostic oncologists therapeutic applications.

Language: Английский

Citations

30
Artificial intelligence for radiological paediatric fracture assessment: a systematic review DOI Creative Commons
Susan C. Shelmerdine, Richard White, Hantao Liu

et al.

Insights into Imaging, Journal Year: 2022, Volume and Issue: 13(1)

Published: June 3, 2022

Majority of research and commercial efforts have focussed on use artificial intelligence (AI) for fracture detection in adults, despite the greater long-term clinical medicolegal implications missed fractures children. The objective this study was to assess available literature regarding diagnostic performance AI tools paediatric assessment imaging, where available, how compares with human readers.MEDLINE, Embase Cochrane Library databases were queried studies published between 1 January 2011 2021 using terms related 'fracture', 'artificial intelligence', 'imaging' 'children'. Risk bias assessed a modified QUADAS-2 tool. Descriptive statistics accuracies collated.Nine eligible articles from 362 publications included, most (8/9) evaluating radiographs, elbow being common body part. Nearly all used data derived single institution, deep learning methodology only few (2/9) performing external validation. Accuracy rates generated by ranged 88.8 97.9%. In two three compared readers, sensitivity marginally higher, but not statistically significant.Wide heterogeneity limited information algorithm datasets makes it difficult understand such may generalise wider population. Further multicentric dataset real-world evaluation would help better impact these tools.

Language: Английский

Citations

30