Deep Learning with Limited Data: Organ Segmentation Performance by U-Net
(1) Background: The effectiveness of deep learning artificial intelligence depends on data availability, often requiring large volumes of data to effectively train an algorithm. However, few studies have explored the minimum number of images needed for optimal algorithmic performance. (2) Methods: T...
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2020-07-01
|
| Series: | Electronics |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2079-9292/9/8/1199 |
| id |
doaj-1d88780a7b264907a4a3abbb9e42da60 |
|---|---|
| record_format |
Article |
| spelling |
doaj-1d88780a7b264907a4a3abbb9e42da602020-11-25T03:32:07ZengMDPI AGElectronics2079-92922020-07-0191199119910.3390/electronics9081199Deep Learning with Limited Data: Organ Segmentation Performance by U-NetMichelle Bardis0Roozbeh Houshyar1Chanon Chantaduly2Alexander Ushinsky3Justin Glavis-Bloom4Madeleine Shaver5Daniel Chow6Edward Uchio7Peter Chang8Department of Radiological Sciences, University of California, Irvine, CA 92617, USADepartment of Radiological Sciences, University of California, Irvine, CA 92617, USADepartment of Radiological Sciences, University of California, Irvine, CA 92617, USAMallinckrodt Institute of Radiology, Washington University Saint Louis, St. Louis, MO 63110, USADepartment of Radiological Sciences, University of California, Irvine, CA 92617, USADepartment of Radiological Sciences, University of California, Irvine, CA 92617, USADepartment of Radiological Sciences, University of California, Irvine, CA 92617, USADepartment of Urology, University of California, Orange, CA 92868, USADepartment of Radiological Sciences, University of California, Irvine, CA 92617, USA(1) Background: The effectiveness of deep learning artificial intelligence depends on data availability, often requiring large volumes of data to effectively train an algorithm. However, few studies have explored the minimum number of images needed for optimal algorithmic performance. (2) Methods: This institutional review board (IRB)-approved retrospective review included patients who received prostate magnetic resonance imaging (MRI) between September 2014 and August 2018 and a magnetic resonance imaging (MRI) fusion transrectal biopsy. T2-weighted images were manually segmented by a board-certified abdominal radiologist. Segmented images were trained on a deep learning network with the following case numbers: 8, 16, 24, 32, 40, 80, 120, 160, 200, 240, 280, and 320. (3) Results: Our deep learning network’s performance was assessed with a Dice score, which measures overlap between the radiologist’s segmentations and deep learning-generated segmentations and ranges from 0 (no overlap) to 1 (perfect overlap). Our algorithm’s Dice score started at 0.424 with 8 cases and improved to 0.858 with 160 cases. After 160 cases, the Dice increased to 0.867 with 320 cases. (4) Conclusions: Our deep learning network for prostate segmentation produced the highest overall Dice score with 320 training cases. Performance improved notably from training sizes of 8 to 120, then plateaued with minimal improvement at training case size above 160. Other studies utilizing comparable network architectures may have similar plateaus, suggesting suitable results may be obtainable with small datasets.https://www.mdpi.com/2079-9292/9/8/1199training sizedeep learningconvolutional neural networkU-Netsegmentationartificial intelligence |
| collection |
DOAJ |
| language |
English |
| format |
Article |
| sources |
DOAJ |
| author |
Michelle Bardis Roozbeh Houshyar Chanon Chantaduly Alexander Ushinsky Justin Glavis-Bloom Madeleine Shaver Daniel Chow Edward Uchio Peter Chang |
| spellingShingle |
Michelle Bardis Roozbeh Houshyar Chanon Chantaduly Alexander Ushinsky Justin Glavis-Bloom Madeleine Shaver Daniel Chow Edward Uchio Peter Chang Deep Learning with Limited Data: Organ Segmentation Performance by U-Net Electronics training size deep learning convolutional neural network U-Net segmentation artificial intelligence |
| author_facet |
Michelle Bardis Roozbeh Houshyar Chanon Chantaduly Alexander Ushinsky Justin Glavis-Bloom Madeleine Shaver Daniel Chow Edward Uchio Peter Chang |
| author_sort |
Michelle Bardis |
| title |
Deep Learning with Limited Data: Organ Segmentation Performance by U-Net |
| title_short |
Deep Learning with Limited Data: Organ Segmentation Performance by U-Net |
| title_full |
Deep Learning with Limited Data: Organ Segmentation Performance by U-Net |
| title_fullStr |
Deep Learning with Limited Data: Organ Segmentation Performance by U-Net |
| title_full_unstemmed |
Deep Learning with Limited Data: Organ Segmentation Performance by U-Net |
| title_sort |
deep learning with limited data: organ segmentation performance by u-net |
| publisher |
MDPI AG |
| series |
Electronics |
| issn |
2079-9292 |
| publishDate |
2020-07-01 |
| description |
(1) Background: The effectiveness of deep learning artificial intelligence depends on data availability, often requiring large volumes of data to effectively train an algorithm. However, few studies have explored the minimum number of images needed for optimal algorithmic performance. (2) Methods: This institutional review board (IRB)-approved retrospective review included patients who received prostate magnetic resonance imaging (MRI) between September 2014 and August 2018 and a magnetic resonance imaging (MRI) fusion transrectal biopsy. T2-weighted images were manually segmented by a board-certified abdominal radiologist. Segmented images were trained on a deep learning network with the following case numbers: 8, 16, 24, 32, 40, 80, 120, 160, 200, 240, 280, and 320. (3) Results: Our deep learning network’s performance was assessed with a Dice score, which measures overlap between the radiologist’s segmentations and deep learning-generated segmentations and ranges from 0 (no overlap) to 1 (perfect overlap). Our algorithm’s Dice score started at 0.424 with 8 cases and improved to 0.858 with 160 cases. After 160 cases, the Dice increased to 0.867 with 320 cases. (4) Conclusions: Our deep learning network for prostate segmentation produced the highest overall Dice score with 320 training cases. Performance improved notably from training sizes of 8 to 120, then plateaued with minimal improvement at training case size above 160. Other studies utilizing comparable network architectures may have similar plateaus, suggesting suitable results may be obtainable with small datasets. |
| topic |
training size deep learning convolutional neural network U-Net segmentation artificial intelligence |
| url |
https://www.mdpi.com/2079-9292/9/8/1199 |
| work_keys_str_mv |
AT michellebardis deeplearningwithlimiteddataorgansegmentationperformancebyunet AT roozbehhoushyar deeplearningwithlimiteddataorgansegmentationperformancebyunet AT chanonchantaduly deeplearningwithlimiteddataorgansegmentationperformancebyunet AT alexanderushinsky deeplearningwithlimiteddataorgansegmentationperformancebyunet AT justinglavisbloom deeplearningwithlimiteddataorgansegmentationperformancebyunet AT madeleineshaver deeplearningwithlimiteddataorgansegmentationperformancebyunet AT danielchow deeplearningwithlimiteddataorgansegmentationperformancebyunet AT edwarduchio deeplearningwithlimiteddataorgansegmentationperformancebyunet AT peterchang deeplearningwithlimiteddataorgansegmentationperformancebyunet |
| _version_ |
1724569573186863104 |