Auto-segmentation of pancreatic tumor in multi-parametric MRI using deep convolutional neural networks

Liang, Ying; Schött, D.; Zhang, Ying; Wang, Zhiwu; Nasief, Haidy; Paulson, E.S.; Hall, William A.; Knechtges, Paul; Erickson, Beth; Li, X. Allen

doi:10.1016/j.radonc.2020.01.021

Cited by 68 publications

(51 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One major contributor to the rapid development of oART has been the increased use of artificial intelligence (AI) [14]. Applications of AI in this field include automated organ segmentation [15][16][17], the use of synthetic image data [18,19] as well as automated treatment planning [20][21][22]. The purpose of this project was to describe the clinical implementation of a commercial solution for CBCT-guided and AIdriven oART.…”

Section: Introductionmentioning

confidence: 99%

Clinical implementation of artificial intelligence-driven cone-beam computed tomography-guided online adaptive radiotherapy in the pelvic region

Sibolt

Andersson

Calmels

et al. 2021

Physics and Imaging in Radiation Oncology

141

139

View full text Add to dashboard Cite

Background and purpose Studies have demonstrated the potential of online adaptive radiotherapy (oART). However, routine use has been limited due to resource demanding solutions. This study reports on experiences with oART in the pelvic region using a novel cone-beam computed tomography (CBCT)-based, artificial intelligence (AI)-driven solution. Material and methods Automated pre-treatment planning for thirty-nine pelvic cases (bladder, rectum, anal, and prostate), and one hundred oART simulations were conducted in a pre-clinical release of Ethos (Varian Medical Systems, Palo Alto, CA). Plan quality, AI-segmentation accuracy, oART feasibility and an integrated calculation-based quality assurance solution were evaluated. Experiences from the first five clinical oART patients (three bladder, one rectum and one sarcoma) are reported. Results Auto-generated pre-treatment plans demonstrated similar planning target volume (PTV) coverage and organs at risk doses, compared to institution reference. More than 75% of AI-segmentations during simulated oART required none or minor editing and the adapted plan was superior in 88% of cases. Limitations in AI-segmentation correlated to cases where AI model training was lacking. The five first treated patients complied well with the median adaptive procedure duration of 17.6 min (from CBCT acceptance to treatment delivery start). The treated bladder patients demonstrated a 42% median primary PTV reduction, indicating a 24%-30% reduction in V 45Gy to the bowel cavity, compared to non-ART. Conclusions A novel commercial oART solution was demonstrated feasible for various pelvic sites. Clinically acceptable AI-segmentation and auto-planning enabled adaptation within reasonable timeslots. Possibilities for reduced PTVs observed for bladder cancer indicated potential for toxicity reductions.

show abstract

Section: Introductionmentioning

confidence: 99%

Clinical implementation of artificial intelligence-driven cone-beam computed tomography-guided online adaptive radiotherapy in the pelvic region

Sibolt

Andersson

Calmels

et al. 2021

Physics and Imaging in Radiation Oncology

141

139

View full text Add to dashboard Cite

show abstract

“…This indicates the NE T2W-FS images could be a potential substitute for CE-T1W images when using CNN for the automatic delineation of primary NPC. Most previous studies on CNN-based primary tumor delineation have relied on the CE MRI [17][18][19][33][34][35], but our results are encouraging and suggest that future adaptations of CNN for NE sequence are warranted to facilitate the reduction of contrast administration for MRI where possible.…”

Section: Discussionmentioning

confidence: 68%

“…With regard to the performance of the CNN primary NPC delineation using CE imaging, our results (DSC, 0.71; ASD, 2.1 mm), are better than those previously reported for U-Net and similar or slightly worse than those reported for using a customized CNN. We used U-Net as our testing reference because it is one of the most general and representative 2D delineation CNN architectures and its encoder-decoder design is the back-bone of many proposed delineation CNN architectures [15,18,[34][35][36][37][38]. Only one other NPC study used the U-Net for primary tumor delineation and reported a DSC of 0.59 and ASD of 6 mm [18] on CE images.…”

Section: Discussionmentioning

confidence: 99%

Non contrast-enhanced imaging as a replacement for contrast-enhanced imaging for MRI automatic delineation of nasopharyngeal carcinoma

Wong

et al. 2020

Preprint

View full text Add to dashboard Cite

Convolutional neural networks (CNNs) show potential for delineating cancers on contrast-enhanced MRI. However, there is world-wide interest in reducing the administration of MRI contrast agents. We aim to determine if CNNs can automatically delineate primary nasopharyngeal carcinoma (NPC) using the non contrast-enhanced (NE) T2-weighted fat-suppressed T1-weighted (CE-T1W) sequence. We e retrospectively analyzed primary tumors in 201 patients with NPC. Six patients were randomly sampled as the training-validation group to avoid over-fitting, and the remaining 195 patients underwent validation analysis. We trained and tested a well-established two-dimensional CNN, U-Net, for tumor delineation on CE-T1W and T2W-FS sequences. CNN-derived delineations on CE-T1W and T2W-FS were compared with manual delineation using the dice similarity coefficient (DSC) and average surface distance (ASD). Differences in DSC and ASD of CNN-derived delineations between CE-T1W and T2W-FS sequences were compared using the Wilcoxon rank test. CNN-derived primary tumor volumes (PTVs) on CE-T1W and T2W-FS were also compared with manual delineation using the Wilcoxon rank test. The CNN's tumor delineation performance on CE-T1W and T2W-FS showed no differences in DSC (0.71±0.09 vs. 0.71±0.09, p=0.50) and ASD (0.21±0.48cm vs. 0.17±0.19cm, p=0.34). The CNN-derived PTVs were larger than those from manual delineation on both CE-T1W (26.3±25.5cm3 vs. 23.5±26.6cm3, p<0.001) and T2W-FS (24.2±23.7cm3 vs. 23.2±26.2 cm3, p<0.001). In conclusion, CNN can automatically delineate primary NPC using the NE T2W-FS sequence which has the potential to be a substitute for the CE-T1W sequence.

show abstract

“…Another limitation of this experience is represented by the lack of an analysis related to the reproducibility of the radiomic features, which represents a necessary step for the clinical implementation in studies based on manual segmentation processes. In this context, a great evolution is expected from automatic contouring systems that are able to ensure precise and standardised contours in a very short time: the feasibility of these approaches on MRI has been recently demonstrated by some experiences, even focused on the abdominal district [53,54].…”

Section: Discussionmentioning

confidence: 99%

Delta Radiomics Analysis for Local Control Prediction in Pancreatic Cancer Patients Treated Using Magnetic Resonance Guided Radiotherapy

et al. 2021

View full text Add to dashboard Cite

The aim of this study is to investigate the role of Delta Radiomics analysis in the prediction of one-year local control (1yLC) in patients affected by locally advanced pancreatic cancer (LAPC) and treated using Magnetic Resonance guided Radiotherapy (MRgRT). A total of 35 patients from two institutions were enrolled: A 0.35 Tesla T2*/T1 MR image was acquired for each case during simulation and on each treatment fraction. Physical dose was converted in biologically effective dose (BED) to compensate for different radiotherapy schemes. Delta Radiomics analysis was performed considering the gross tumour volume (GTV) delineated on MR images acquired at BED of 20, 40, and 60 Gy. The performance of the delta features in predicting 1yLC was investigated in terms of Wilcoxon Mann–Whitney test and area under receiver operating characteristic (ROC) curve (AUC). The most significant feature in predicting 1yLC was the variation of cluster shade calculated at BED = 40 Gy, with a p-value of 0.005 and an AUC of 0.78 (0.61–0.94). Delta Radiomics analysis on low-field MR images might play a promising role in 1yLC prediction for LAPC patients: further studies including an external validation dataset and a larger cohort of patients are recommended to confirm the validity of this preliminary experience.

show abstract

Auto-segmentation of pancreatic tumor in multi-parametric MRI using deep convolutional neural networks

Cited by 68 publications

References 26 publications

Clinical implementation of artificial intelligence-driven cone-beam computed tomography-guided online adaptive radiotherapy in the pelvic region

Clinical implementation of artificial intelligence-driven cone-beam computed tomography-guided online adaptive radiotherapy in the pelvic region

Non contrast-enhanced imaging as a replacement for contrast-enhanced imaging for MRI automatic delineation of nasopharyngeal carcinoma

Delta Radiomics Analysis for Local Control Prediction in Pancreatic Cancer Patients Treated Using Magnetic Resonance Guided Radiotherapy

Contact Info

Product

Resources

About