A convolutional neural network algorithm for automatic segmentation of head and neck organs at risk using deep lifelong learning

Chan, Jason W.; Kearney, Vasant; Haaf, Samuel; Wu, Susan; Bogdanov, Madeleine; Reddick, Mariah; Dixit, Nayha; Sudhyadhom, Atchar; Chen, Josephine; Yom, Sue S.; Solberg, Timothy D.

doi:10.1002/mp.13495

Cited by 55 publications

(37 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Parotid glands DC (%) 92 AE 4, 37 91 AE 2, 75 88 AE 2, 46 91 m;f (N), 45 88, 53 87 AE 3(N), 60 87 AE 4 (N), 24 87, 64 86 AE 2 (N), 40 86 AE 3, 48 86 AE 4, 24 86 AE 5 (N), 31 86 AE 5, 42 86 AE 5, 40 86 AE 7, 93 91 m;f , 72 85 AE 2, 83 85 AE 3, 26 85 AE 4, 91 85 AE 4, 30 85 AE 5, 47 91 m;f (DL), 29 85, 60 84 AE 3, 34 84 AE 3 (N), 55 84 AE 4 (•), 60 84 AE 7 (N,IM), 66 84, 76 91 m;f , 22 91 m;f , 23 83 AE 2, 50 83 AE 3, 58 83 AE 5 (•), 36 83 AE 5, 86 83 AE 6, 36 83 AE 6 (N), 56 91 m;f , 95 91 m;f , 81 AE 4 (N), 70 81 AE 5, 28 81 AE 8 (N), 49 81 AE 8, 27 81 (N), 54 91 m;f , 52 91 m;f (ABAS), 29 79 (MR), 68 91 m;f , 77 79, 87 79, 57 77 AE 6, 65 91 m;f (N), 69 91 m;f (N), 35 76 AE 6, 63 76 (CT), 68 91 m;f , 35 75, 51 72 AE 10, …”

Section: Resultsmentioning

confidence: 99%

“…Submandibular glands DC (%) 91 m;f , 22 85 AE 10, 42 85, 60 84 AE 6, 24 83, 53 82 AE 5, 86 82 AE 5 (N), 40 82 AE 7, 30 82 AE 7, 50 81 AE 4, 46 81 AE 6 (N), 55 80 AE 7 (N), 24 80 AE 7, 26 80 AE 8 (•), 60 91 m;f , 77 91 m;f , 23 78 AE 7 (N), 60 78 AE 8 (N,IM), 66 77 AE 6, 34 75 AE 13 (N), 31 VC (%) TPR: 87 AE 5, 24 85 AE 6 (N), 55 80 AE 11, 50 79 AE 8 (N), 24 79 AE 9 (N), 40 72 AE 16 (N) 31 ; PPV: 85 AE 9 (N), 40 83 AE 11, 24 82 AE 9 (N), 24 82 AE 11 (N), 31 ASD (mm) ASSD: 1.2, 53 1.3 AE 1.2 26 ; ASD91 m;f : 91 m;f (N,IM) 66 ; ASD91 m;f : 0.9 AE 0.5 (N), 55 1.2 AE 0.7, 24 1.4 AE 1.0 (N), 40 2.0 AE 1.9 (N) 24 ; DTA91 m;f : 91 m;f , 77 1.2 AE 1.3, 42 1.9 AE 1.4 87 Brainstem DC (%) 93 AE 1, 97 93 AE 3, 27 92 AE 3, 40 92, 53 91 AE 1, 86 91 AE 3, 43 90 AE 1, 26 90 AE 2,…”

Section: Resultsmentioning

confidence: 99%

“…Parotid glands [22][23][24][26][27][28][29][30][31][32][34][35][36][37]40,42,[45][46][47][48][49][50][51][52][53][54][55][56][57][58]60,[63][64][65][66][68][69][70]72,73,[75][76][77][78]80,[82][83][84]86,87,90,91,93,95…”

Section: Organ At Riskunclassified

“…Average distance to agreement 27,42,68,77,84,87 Legend: |A| and |B| are the number of voxels in volumetric masks A (e.g., ground truth) and B (e.g., auto-segmentation), respectively, and |@A| and |@B| are the number of voxels in the corresponding subsets of surface voxels @A and @B, respectively. The Euclidean distances of voxels a and b to surfaces @B and @A, respectively, are defined as dða; @BÞ ¼ min b2@B kax02010; bk and dðb; @AÞ ¼ min a2@A kbx02010; ak, respectively.…”

Section: Dta Avgmentioning

confidence: 99%

“…, 2.0 AE 1.9 (PCM)87 Cervical esophagus with the cricopharyngeal inlet, upper esophageal sphincter (UES) DC (%) 86 AE 342 , 82 AE 6 28 , 81 AE 14 (UES) 50 , 81 AE 7 36 , 70 AE 7 61 , 69 AE 10 26 , 62 51 , 60 AE 11 50 , 91 m;f 52 , 91 m;f 23 , 35 53 VC (%) TPR: 80 AE 16 (UES) 50 , 50 AE 15 50 ; FDR: 15 AE 14 (UES) 50 , 21 AE 14 50 AE 4 47 , 86 AE 4 65 , 86 AE 7 42 , 83 AE 8 28 , 80 AE 5 40 , 78 AE 4 50 , 91 m;f 52 , 77 AE 7 26 , 74 51 , 91 m;f 35 , 71 53 , 91 m;f 77 AE 10 60 , 82 AE 7 (•) 60 , 74 53 , 66 AE 13 36 , 65 AE 7 26 , 41 AE 8 (•) 36 , 91 m;f 77…”

mentioning

confidence: 99%

See 4 more Smart Citations

Auto‐segmentation of organs at risk for head and neck radiotherapy planning: From atlas‐based to deep learning methods

et al. 2020

View full text Add to dashboard Cite

Radiotherapy (RT) is one of the basic treatment modalities for cancer of the head and neck (H&N), which requires a precise spatial description of the target volumes and organs at risk (OARs) to deliver a highly conformal radiation dose to the tumor cells while sparing the healthy tissues. For this purpose, target volumes and OARs have to be delineated and segmented from medical images. As manual delineation is a tedious and time-consuming task subjected to intra/interobserver variability, computerized auto-segmentation has been developed as an alternative. The field of medical imaging and RT planning has experienced an increased interest in the past decade, with new emerging trends that shifted the field of H&N OAR auto-segmentation from atlas-based to deep learning-based approaches. In this review, we systematically analyzed 78 relevant publications on auto-segmentation of OARs in the H&N region from 2008 to date, and provided critical discussions and recommendations from various perspectives: image modalityboth computed tomography and magnetic resonance image modalities are being exploited, but the potential of the latter should be explored more in the future; OARthe spinal cord, brainstem, and major salivary glands are the most studied OARs, but additional experiments should be conducted for several less studied soft tissue structures; image databaseseveral image databases with the corresponding ground truth are currently available for methodology evaluation, but should be augmented with data from multiple observers and multiple institutions; methodologycurrent methods have shifted from atlas-based to deep learning autosegmentation, which is expected to become even more sophisticated; ground truthdelineation guidelines should be followed and participation of multiple experts from multiple institutions is recommended; performance metricsthe Dice coefficient as the standard volumetric overlap metrics should be accompanied with at least one distance metrics, and combined with clinical acceptability scores and risk assessments; segmentation performancethe best performing methods achieve clinically acceptable auto-segmentation for several OARs, however, the dosimetric impact should be also studied to provide clinically relevant endpoints for RT planning.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Organ At Riskunclassified

Section: Dta Avgmentioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations

Auto‐segmentation of organs at risk for head and neck radiotherapy planning: From atlas‐based to deep learning methods

et al. 2020

View full text Add to dashboard Cite

show abstract

Evaluation of deep learning‐based auto‐segmentation algorithms for delineating clinical target volume and organs at risk involving data for 125 cervical cancer patients

Zhi

Chang

Zhao

et al. 2020

J Applied Clin Med Phys

View full text Add to dashboard Cite

Objective: To evaluate the accuracy of a deep learning-based auto-segmentation mode to that of manual contouring by one medical resident, where both entities tried to mimic the delineation "habits" of the same clinical senior physician. Methods: This study included 125 cervical cancer patients whose clinical target volumes (CTVs) and organs at risk (OARs) were delineated by the same senior physician. Of these 125 cases, 100 were used for model training and the remaining 25 for model testing. In addition, the medical resident instructed by the senior physician for approximately 8 months delineated the CTVs and OARs for the testing cases. The dice similarity coefficient (DSC) and the Hausdorff Distance (HD) were used to evaluate the delineation accuracy for CTV, bladder, rectum, small intestine, femoral-head-left, and femoral-head-right. Results: The DSC values of the auto-segmentation model and manual contouring by the resident were, respectively, 0.86 and 0.83 for the CTV (P < 0.05), 0.91 and 0.91 for the bladder (P > 0.05), 0.88 and 0.84 for the femoral-head-right (P < 0.05), 0.88 and 0.84 for the femoral-head-left (P < 0.05), 0.86 and 0.81 for the small intestine (P < 0.05), and 0.81 and 0.84 for the rectum (P > 0.05). The HD (mm) values were, respectively, 14.84 and 18.37 for the CTV (P < 0.05), 7.82 and 7.63 for the bladder (P > 0.05), 6.18 and 6.75 for the femoral-head-right (P > 0.05), 6.17 and 6.31 for the femoral-head-left (P > 0.05), 22.21 and 26.70 for the small intestine (P > 0.05), and 7.04 and 6.13 for the rectum (P > 0.05). The auto-segmentation model took approximately 2 min to delineate the CTV and OARs while the resident took approximately 90 min to complete the same task. Conclusion: The auto-segmentation model was as accurate as the medical resident but with much better efficiency in this study. Furthermore, the auto-segmentation approach offers additional perceivable advantages of being consistent and ever improving when compared with manual approaches.

show abstract

Improved contrast and noise of megavoltage computed tomography (MVCT) through cycle‐consistent generative machine learning

et al. 2020

Self Cite

View full text Add to dashboard Cite

Purpose: Megavoltage computed tomography (MVCT) has been implemented on many radiation therapy treatment machines as a tomographic imaging modality that allows for three-dimensional visualization and localization of patient anatomy. Yet MVCT images exhibit lower contrast and greater noise than its kilovoltage CT (kVCT) counterpart. In this work, we sought to improve these disadvantages of MVCT images through an image-to-image-based machine learning transformation of MVCT and kVCT images. We demonstrated that by learning the style of kVCT images, MVCT images can be converted into high-quality synthetic kVCT (skVCT) images with higher contrast and lower noise, when compared to the original MVCT. Methods: Kilovoltage CT and MVCT images of 120 head and neck (H&N) cancer patients treated on an Accuray TomoHD system were retrospectively analyzed in this study. A cycle-consistent generative adversarial network (CycleGAN) machine learning, a variant of the generative adversarial network (GAN), was used to learn Hounsfield Unit (HU) transformations from MVCT to kVCT images, creating skVCT images. A formal mathematical proof is given describing the interplay between function sensitivity and input noise and how it applies to the error variance of a high-capacity function trained with noisy input data. Finally, we show how skVCT shares distributional similarity to kVCT for various macro-structures found in the body. Results: Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were improved in skVCT images relative to the original MVCT images and were consistent with kVCT images. Specifically, skVCT CNR for muscle-fat, bone-fat, and bone-muscle improved to 14.8 AE 0.4, 122.7 AE 22.6, and 107.9 AE 22.4 compared with 1.6 AE 0.3, 7.6 AE 1.9, and 6.0 AE 1.7, respectively, in the original MVCT images and was more consistent with kVCT CNR values of 15.2 AE 0.8, 124.9 AE 27.0, and 109.7 AE 26.5, respectively. Noise was significantly reduced in skVCT images with SNR values improving by roughly an order of magnitude and consistent with kVCT SNR values. Axial slice mean (S-ME) and mean absolute error (S-MAE) agreement between kVCT and MVCT/skVCT improved, on average, from −16.0 and 109.1 HU to 8.4 and 76.9 HU, respectively. Conclusions: A kVCT-like qualitative aid was generated from input MVCT data through a Cycle-GAN instance. This qualitative aid, skVCT, was robust toward embedded metallic material, dramatically improves HU alignment from MVCT, and appears perceptually similar to kVCT with SNR and CNR values equivalent to that of kVCT images.

show abstract

A convolutional neural network algorithm for automatic segmentation of head and neck organs at risk using deep lifelong learning

Cited by 55 publications

References 43 publications

Auto‐segmentation of organs at risk for head and neck radiotherapy planning: From atlas‐based to deep learning methods

Auto‐segmentation of organs at risk for head and neck radiotherapy planning: From atlas‐based to deep learning methods

Evaluation of deep learning‐based auto‐segmentation algorithms for delineating clinical target volume and organs at risk involving data for 125 cervical cancer patients

Improved contrast and noise of megavoltage computed tomography (MVCT) through cycle‐consistent generative machine learning

Contact Info

Product

Resources

About