國立高雄大學圖資館 |

Language: English

Back

Cats to CATs with RadImageNet: A Transformative Platform for Medical Imaging AI Research.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Cats to CATs with RadImageNet: A Transformative Platform for Medical Imaging AI Research.
Author:	Mei, Xueyan.
Published:	Ann Arbor : ProQuest Dissertations & Theses, 2022
Description:	154 p.
Notes:	Source: Dissertations Abstracts International, Volume: 83-07, Section: B.
Notes:	Advisor: Fayad, Zahi A.;Yang, Yang.
Contained By:	Dissertations Abstracts International83-07B.
Subject:	Biomedical engineering.
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28869171
ISBN:	9798759991564

Cats to CATs with RadImageNet: A Transformative Platform for Medical Imaging AI Research.
Mei, Xueyan.

Cats to CATs with RadImageNet: A Transformative Platform for Medical Imaging AI Research. - Ann Arbor : ProQuest Dissertations & Theses, 2022 - 154 p.

Source: Dissertations Abstracts International, Volume: 83-07, Section: B.

Thesis (Ph.D.)--Icahn School of Medicine at Mount Sinai, 2022.

This item must not be sold to any third party vendors.

Most current medical imaging Artificial Intelligence (AI) relies upon transfer learning using convolutional neural networks (CNNs) developed upon ImageNet, a large database of natural world images, including cats, dogs, and vehicles. Size, diversity, and similarity of the source data determine the success of the transfer learning on the target data. ImageNet is large and diverse, but there is a significant dissimilarity between its natural world images and medical images. Despite low similarity, ImageNet pretrained models are widely used in medical image classification. In chapter 2, ImageNet-based ResNet18 model was used to diagnose COVID-19 patients based on chest CT scans and associated clinical information and this model achieved senior radiologist level in recognition of COVID-19 patients. However, a medical imaging-only dataset similar to ImageNet and pretrained models containing only radiologic features could improve model performance. In this thesis, we curated a standardized database, RadImageNet, consisting of 1.35 million annotated medical images consisting of CT, MRI, and ultrasound of musculoskeletal, neurologic, oncologic, gastrointestinal, endocrine, abdominal, and pulmonary pathologies from over 130,000 patients. This database is unprecedented in scale and breadth in the medical imaging field that allows for training convolutional neural networks from scratch without importing weights from existing pre-trained models and the models derived from the RadImageNet database could be a better starting point to improve medical imaging applications that require transfer learning. The establishment of the RadImageNet database and associated pretrained models as well as comparisons to ImageNet models on eight independent applications including datasets used in chapter 2 were presented in chapter 3. To further study inner connections between different modalities from the RadImageNet database, we developed RadImageNet modality and anatomy models and created machine learning classifiers to select best parameter settings. Overall, this thesis evaluated ImageNet models and RadImageNet models on multiple medical imaging datasets and RadImageNet derived models and selection of CNN parameters could facilitate medical imaging AI research in recognition rate and saving turnaround time.

ISBN: 9798759991564Subjects--Topical Terms:

190330
Biomedical engineering.
Subjects--Index Terms:

Biomedical images

Cats to CATs with RadImageNet: A Transformative Platform for Medical Imaging AI Research.
LDR:03470nmm a2200349 4500 001 616508
005 20220513114356.5
008 220920s2022 ||||||||||||||||| ||eng d
020 $a 9798759991564
035 $a (MiAaPQ)AAI28869171
035 $a AAI28869171
040 $a MiAaPQ $c MiAaPQ
100 1 $a Mei, Xueyan. $3 915889
245 1 0 $a Cats to CATs with RadImageNet: A Transformative Platform for Medical Imaging AI Research.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2022
300 $a 154 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-07, Section: B.
500 $a Advisor: Fayad, Zahi A.;Yang, Yang.
502 $a Thesis (Ph.D.)--Icahn School of Medicine at Mount Sinai, 2022.
506 $a This item must not be sold to any third party vendors.
520 $a Most current medical imaging Artificial Intelligence (AI) relies upon transfer learning using convolutional neural networks (CNNs) developed upon ImageNet, a large database of natural world images, including cats, dogs, and vehicles. Size, diversity, and similarity of the source data determine the success of the transfer learning on the target data. ImageNet is large and diverse, but there is a significant dissimilarity between its natural world images and medical images. Despite low similarity, ImageNet pretrained models are widely used in medical image classification. In chapter 2, ImageNet-based ResNet18 model was used to diagnose COVID-19 patients based on chest CT scans and associated clinical information and this model achieved senior radiologist level in recognition of COVID-19 patients. However, a medical imaging-only dataset similar to ImageNet and pretrained models containing only radiologic features could improve model performance. In this thesis, we curated a standardized database, RadImageNet, consisting of 1.35 million annotated medical images consisting of CT, MRI, and ultrasound of musculoskeletal, neurologic, oncologic, gastrointestinal, endocrine, abdominal, and pulmonary pathologies from over 130,000 patients. This database is unprecedented in scale and breadth in the medical imaging field that allows for training convolutional neural networks from scratch without importing weights from existing pre-trained models and the models derived from the RadImageNet database could be a better starting point to improve medical imaging applications that require transfer learning. The establishment of the RadImageNet database and associated pretrained models as well as comparisons to ImageNet models on eight independent applications including datasets used in chapter 2 were presented in chapter 3. To further study inner connections between different modalities from the RadImageNet database, we developed RadImageNet modality and anatomy models and created machine learning classifiers to select best parameter settings. Overall, this thesis evaluated ImageNet models and RadImageNet models on multiple medical imaging datasets and RadImageNet derived models and selection of CNN parameters could facilitate medical imaging AI research in recognition rate and saving turnaround time.
590 $a School code: 1734.
650 4 $a Biomedical engineering. $3 190330
650 4 $a Artificial intelligence. $3 194058
653 $a Biomedical images
653 $a Deep learning
653 $a Machine learning
653 $a Transfer learning
690 $a 0541
690 $a 0800
710 2 $a Icahn School of Medicine at Mount Sinai. $b Pharmacology and System Biology. $3 915890
773 0 $t Dissertations Abstracts International $g 83-07B.
790 $a 1734
791 $a Ph.D.
792 $a 2022
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28869171