- Email: [email protected]
Computer-aided diagnosis (CAD) and image-guided decision support
Computerized Medical Imaging and Graphics 31 (2007) 195–197
Editorial
Computer-aided diagnosis (CAD) and image-guided decision support
Computer-aided diagnosis (CAD) research started in the early 1980s and has gradually evolved as a clinically supported tool. In mammography, CAD has, in fact, become a part of the routine clinical operation for detection of breast cancers in many medical centers and screening sites in the United States. In addition, various CAD schemes are being developed for detection and classification of many different kinds of lesions obtained with the use of various imaging modalities, because the concept of CAD is broad and general in assisting radiologists by providing the computer output as a “second opinion”. The usefulness and practicality of CAD, however, depend on many factors, including the availability of digital image data, computer power, and high-quality display and image-archiving systems. Therefore, it is apparent that CAD needs to be integrated into a part of PACS in the future. Image-based knowledge discovery and decision support by use of CAD are a new trend in research that translates CAD diagnostic results to assist in short- and longterm treatments. PACS was developed more than 20 years ago and has become an integral part of daily clinical operations. Integration of CAD with PACS would take advantage of the image resources in PACS and enhance the value of CAD. Because of this, CAD applications have emerged into mainstream imageaided clinical practice. This special issue considers the current status and future potential of CAD applications, knowledge discovery and decision support methods, as well as their integration with PACS. This special issue on computer-aided diagnosis (CAD), image-guided decision support, and DICOM-based Integration includes 15 papers and is loosely categorized into six major topics: CAD overview, CAD research and development trends, some targeted CAD applications, decision support, CAD systems and evaluation methodology, and security and system integration considerations. 1. Overview Doi—computer-aided diagnosis in medical imaging: historical review, current status and future potential. Computer-aided diagnosis has become one of the major research subjects in medical imaging and diagnostic radiology. In this article, the motivation and philosophy for early development of CAD schemes are presented together with the current 0895-6111/$ – see front matter © 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.compmedimag.2007.02.001
status and future potential of CAD in a PACS environment. With CAD, radiologists use the computer output as a “second opinion” and make the final decisions. CAD is a concept that was established by taking into account equally the roles of physicians and computers, whereas automated computer diagnosis is a concept based on computer algorithms only. With CAD, the performance by computers does not have to be comparable to or better than that by physicians, but needs to be complementary to that by physicians. 2. CAD research and development trends In this category, five papers are included: • Katsuragawa et al.—Computer-Aided Diagnosis in Chest Radiography; • Nishikawa—Current Status and Future Directions of Computer-Aided Diagnosis in Mammography; • Behrens et al.—Computer Assistance for MR Based Diagnosis of Breast Cancer: Present and Future Challenges; • Li—Recent Progress in Computer-Aided Diagnosis of Lung Nodule on Thin-Section CT; • Kobatake—Future CAD in Multi-Dimensional Medical Images-Project on Multi-Organ, Multi-Disease CAD System. These papers discuss CAD developments during the past years and their future directions in chest radiography, mammography, MR breast imaging, CT nodules, and multi-dimensional medical images, organs, and diseases. For chest radiography, Katsuragawa and Doi describe CAD schemes for detection of lung nodules, interstitial lung diseases, interval changes, and asymmetric opacities, and also for differential diagnosis of lung nodules and interstitial lung diseases. Observer performance studies indicate clearly that radiologists’ diagnostic accuracy was improved significantly when radiologists used a computer output in their interpretation of chest radiographs. By use of chest radiographs, the automated recognition methods for the patient and the projection view may be useful for integrating the chest CAD schemes into PACS. Nishikawa presents the current status and future directions of CAD in mammography. Although CAD schemes have high sensitivity, but poor specificity compared to radiologists, CAD has been shown both
196
Editorial / Computerized Medical Imaging and Graphics 31 (2007) 195–197
in observer studies and in clinical evaluations to help radiologists find additional cancers. Recent clinical studies indicate that CAD increases the number of cancers detected by approximately 10%, which is comparable to double reading by two radiologists. Behrens et al. present a review and future challenges of CAD for MRI of the breast. Dynamic contrast-enhanced MRI of the breast has become a robust and successful method for diagnosis of high-risk cases. With MR spectroscopy and MR-guided biopsies of the breast, the limitations and technical challenges for CAD are discussed regarding registration methods, segmentation issues, as well as morphological and dynamic lesion analysis. Li describes recent progress in the development and evaluation of CAD schemes for detection and characterization of lung nodules on thin-section CT. Observer performance studies have indicated the potential clinical usefulness of CAD schemes. Although current CAD schemes in characterization of nodules in CT would be able to improve radiologists’ performance, current schemes for nodule detection include many false positives, which need to be improved in the future. Kobatake describes an overview and future CAD technologies related to a large research project on “Intelligent Assistance in Diagnosis of Multi-dimensional Medical Images” initiated in Japan in 2003. The objective of this project is to develop a multi-organ, multi-disease CAD system by incorporating both anatomical knowledge of the human body and diagnostic knowledge of various types of diseases. 3. Targeted CAD applications Two papers are included in this category: • Yoshida et al.—CAD in CT Colonography without and with Fecal Tagging: Progress and Challenges; • Chan—Computer Aided Detection of Small Acute Intracranial Hhemorrhage (AIH) on Computer Tomography of Brain. Yoshida and Nappi present the current status and future challenges in CAD for computed tomographic colonography (CTC). Rapid technical developments have advanced CAD for CTC substantially to establish a basic scheme for detection of polyps based on sophisticated three-dimensional processing, analysis, and display techniques. The latest CAD systems provide a clinically acceptable high sensitivity and a low false positive rate. Chan describes a CAD tool, using classical image segmentation algorithms augmented with the knowledge of possible locations of AIH within the brain, to assist general radiologists and physicians in the daily clinical environment and to improve their diagnostic accuracy for AIH. 4. Decision support Decision support means that the CAD method developed contributes to a component of a larger CAD system for achieving a common goal. The goal could be an assessment process or contributing decision support for a treatment plan. Two papers are included in this category:
• Zhang et al.—Automatic Bone Age Assessment for Young Children from Newborn to 7-Year-Old Using Carpal Bones; • Liu et al.—Image-assisted Knowledge Discovery and Decision Support in Radiation Therapy Planning. Zhang presents a CAD component in bone age assessment of children from newborn to 7-year-old based on features of carpal bones in a hand radiograph. The results can be combined with features derived from other bones of the hand. Together, the results contribute to a more comprehensive CAD bone age assessment system of children from newborn to 18-yearold. Liu describes an image-assisted knowledge discovery and decision-support method derived from using the results of various radiation therapy imaging systems, the outcome of which is used for facilitating and improving the accuracy of radiation treatment planning for the patient. 5. CAD systems and evaluation This category includes three papers. The first two are on CAD systems, and the third is a CAD evaluation: • Gertych et al.—Bone Age Assessment of Children using a Digital Hand Atlas; • Brown et al.—CAD in Clinical Trials: Current Role and Architectural Requirements; • Li—Improvement of Bias and Generalizability for ComputerAided Diagnostic Schemes. Gertych presents a CAD system for bone age assessment of children from new born to 18-year-old using a hand radiograph. The system consists of two components: a digital hand atlas with a large database of evenly distributed normal hand radiographs from four races and two genders, and related clinical data; and a CAD application for bone age assessment. The bone age of a given child can be assessed by comparing the child bone features with that of the atlas. Brown describes the role of CAD in clinical trials and the requirements for the CAD system to be used in a successful clinical trial. If a CAD detection system is to become clinically essential in the application to radiation treatment planning, it should be validated in oncologic clinical trials. However, CAD software alone is not sufficient for conducting a clinical trial. There are a number of architectural requirements such as receiving images (from multiple field sites), a fault-tolerance database for storing quantitative measures, and data-mining and reporting capabilities. Li discusses methods for improvement of bias and generalizability of CAD schemes. The reliable evaluation of CAD schemes is as important as the development of such schemes in the field of CAD research. In the past, the performance of CAD schemes has been evaluated by use of various methods such as resubstitution, leave-one-out, cross-validation, and hold-out. This article addresses important issues related to the bias and generalization performance of CAD schemes trained with limited datasets, and analysis of pitfalls in the incorrect use of various methods and approaches to the reduction of bias and variance.
Editorial / Computerized Medical Imaging and Graphics 31 (2007) 195–197
6. Security and integration The integration of CAD and PACS requires consideration of two issues: the method of integration and data security. Two papers by Zhou in this category discuss these topics: • Zhou et al.—CAD-PACS Integration Tool Kit Based on DICOM Secondary Capture, Structured Report and IHE Workflow Profiles; • Zhou—Data Security Assurance in CAD-PACS Integration. In his first paper, Zhou presents a self-contained DICOMbased toolkit based on the screen capture and the structured report of the DICOM standard; and IHE (Integrating Healthcare Enterprise) workflow profiles. His second paper provides a method to insure the integrity of image and CAD output data in the PACS environment by use of a lossless digital signatureembedding method. In summary, in addition to the editorial, the 15 papers included in this special issue on computer-aided diagnosis (CAD) and image-guided decision support covers six categories of state-of-the-art CAD research and development. It ranges
197
from a CAD overview and trends, targeted CAD applications, knowledge discovery and decision support, CAD systems and evaluation, to security and system integration. It is our belief that the integration of CAD with PACS will lead to the emergence of CAD applications into mainstream image-aided clinical practice in both diagnosis and treatment. Kunio Doi ∗ Kurt Rossmann Laboratories for Radiologic Image Research, Department of Radiology, The University of Chicago, 5841 South Maryland Avenue, Chicago, IL 60637, United States H.K. Huang Division of Imaging Informatics, Departments of Radiology and Biomedical Engineering, University of Southern California, 4676 Admiralty Way, Suite 601, Marina del Ray, CA 90292, United States ∗ Corresponding
author. Tel.: +1 773 702 6954; fax: +1 773 702 0371. E-mail addresses: [email protected] (K. Doi), [email protected] (H.K. Huang)
Editorial
Computer-aided diagnosis (CAD) and image-guided decision support
Computer-aided diagnosis (CAD) research started in the early 1980s and has gradually evolved as a clinically supported tool. In mammography, CAD has, in fact, become a part of the routine clinical operation for detection of breast cancers in many medical centers and screening sites in the United States. In addition, various CAD schemes are being developed for detection and classification of many different kinds of lesions obtained with the use of various imaging modalities, because the concept of CAD is broad and general in assisting radiologists by providing the computer output as a “second opinion”. The usefulness and practicality of CAD, however, depend on many factors, including the availability of digital image data, computer power, and high-quality display and image-archiving systems. Therefore, it is apparent that CAD needs to be integrated into a part of PACS in the future. Image-based knowledge discovery and decision support by use of CAD are a new trend in research that translates CAD diagnostic results to assist in short- and longterm treatments. PACS was developed more than 20 years ago and has become an integral part of daily clinical operations. Integration of CAD with PACS would take advantage of the image resources in PACS and enhance the value of CAD. Because of this, CAD applications have emerged into mainstream imageaided clinical practice. This special issue considers the current status and future potential of CAD applications, knowledge discovery and decision support methods, as well as their integration with PACS. This special issue on computer-aided diagnosis (CAD), image-guided decision support, and DICOM-based Integration includes 15 papers and is loosely categorized into six major topics: CAD overview, CAD research and development trends, some targeted CAD applications, decision support, CAD systems and evaluation methodology, and security and system integration considerations. 1. Overview Doi—computer-aided diagnosis in medical imaging: historical review, current status and future potential. Computer-aided diagnosis has become one of the major research subjects in medical imaging and diagnostic radiology. In this article, the motivation and philosophy for early development of CAD schemes are presented together with the current 0895-6111/$ – see front matter © 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.compmedimag.2007.02.001
status and future potential of CAD in a PACS environment. With CAD, radiologists use the computer output as a “second opinion” and make the final decisions. CAD is a concept that was established by taking into account equally the roles of physicians and computers, whereas automated computer diagnosis is a concept based on computer algorithms only. With CAD, the performance by computers does not have to be comparable to or better than that by physicians, but needs to be complementary to that by physicians. 2. CAD research and development trends In this category, five papers are included: • Katsuragawa et al.—Computer-Aided Diagnosis in Chest Radiography; • Nishikawa—Current Status and Future Directions of Computer-Aided Diagnosis in Mammography; • Behrens et al.—Computer Assistance for MR Based Diagnosis of Breast Cancer: Present and Future Challenges; • Li—Recent Progress in Computer-Aided Diagnosis of Lung Nodule on Thin-Section CT; • Kobatake—Future CAD in Multi-Dimensional Medical Images-Project on Multi-Organ, Multi-Disease CAD System. These papers discuss CAD developments during the past years and their future directions in chest radiography, mammography, MR breast imaging, CT nodules, and multi-dimensional medical images, organs, and diseases. For chest radiography, Katsuragawa and Doi describe CAD schemes for detection of lung nodules, interstitial lung diseases, interval changes, and asymmetric opacities, and also for differential diagnosis of lung nodules and interstitial lung diseases. Observer performance studies indicate clearly that radiologists’ diagnostic accuracy was improved significantly when radiologists used a computer output in their interpretation of chest radiographs. By use of chest radiographs, the automated recognition methods for the patient and the projection view may be useful for integrating the chest CAD schemes into PACS. Nishikawa presents the current status and future directions of CAD in mammography. Although CAD schemes have high sensitivity, but poor specificity compared to radiologists, CAD has been shown both
196
Editorial / Computerized Medical Imaging and Graphics 31 (2007) 195–197
in observer studies and in clinical evaluations to help radiologists find additional cancers. Recent clinical studies indicate that CAD increases the number of cancers detected by approximately 10%, which is comparable to double reading by two radiologists. Behrens et al. present a review and future challenges of CAD for MRI of the breast. Dynamic contrast-enhanced MRI of the breast has become a robust and successful method for diagnosis of high-risk cases. With MR spectroscopy and MR-guided biopsies of the breast, the limitations and technical challenges for CAD are discussed regarding registration methods, segmentation issues, as well as morphological and dynamic lesion analysis. Li describes recent progress in the development and evaluation of CAD schemes for detection and characterization of lung nodules on thin-section CT. Observer performance studies have indicated the potential clinical usefulness of CAD schemes. Although current CAD schemes in characterization of nodules in CT would be able to improve radiologists’ performance, current schemes for nodule detection include many false positives, which need to be improved in the future. Kobatake describes an overview and future CAD technologies related to a large research project on “Intelligent Assistance in Diagnosis of Multi-dimensional Medical Images” initiated in Japan in 2003. The objective of this project is to develop a multi-organ, multi-disease CAD system by incorporating both anatomical knowledge of the human body and diagnostic knowledge of various types of diseases. 3. Targeted CAD applications Two papers are included in this category: • Yoshida et al.—CAD in CT Colonography without and with Fecal Tagging: Progress and Challenges; • Chan—Computer Aided Detection of Small Acute Intracranial Hhemorrhage (AIH) on Computer Tomography of Brain. Yoshida and Nappi present the current status and future challenges in CAD for computed tomographic colonography (CTC). Rapid technical developments have advanced CAD for CTC substantially to establish a basic scheme for detection of polyps based on sophisticated three-dimensional processing, analysis, and display techniques. The latest CAD systems provide a clinically acceptable high sensitivity and a low false positive rate. Chan describes a CAD tool, using classical image segmentation algorithms augmented with the knowledge of possible locations of AIH within the brain, to assist general radiologists and physicians in the daily clinical environment and to improve their diagnostic accuracy for AIH. 4. Decision support Decision support means that the CAD method developed contributes to a component of a larger CAD system for achieving a common goal. The goal could be an assessment process or contributing decision support for a treatment plan. Two papers are included in this category:
• Zhang et al.—Automatic Bone Age Assessment for Young Children from Newborn to 7-Year-Old Using Carpal Bones; • Liu et al.—Image-assisted Knowledge Discovery and Decision Support in Radiation Therapy Planning. Zhang presents a CAD component in bone age assessment of children from newborn to 7-year-old based on features of carpal bones in a hand radiograph. The results can be combined with features derived from other bones of the hand. Together, the results contribute to a more comprehensive CAD bone age assessment system of children from newborn to 18-yearold. Liu describes an image-assisted knowledge discovery and decision-support method derived from using the results of various radiation therapy imaging systems, the outcome of which is used for facilitating and improving the accuracy of radiation treatment planning for the patient. 5. CAD systems and evaluation This category includes three papers. The first two are on CAD systems, and the third is a CAD evaluation: • Gertych et al.—Bone Age Assessment of Children using a Digital Hand Atlas; • Brown et al.—CAD in Clinical Trials: Current Role and Architectural Requirements; • Li—Improvement of Bias and Generalizability for ComputerAided Diagnostic Schemes. Gertych presents a CAD system for bone age assessment of children from new born to 18-year-old using a hand radiograph. The system consists of two components: a digital hand atlas with a large database of evenly distributed normal hand radiographs from four races and two genders, and related clinical data; and a CAD application for bone age assessment. The bone age of a given child can be assessed by comparing the child bone features with that of the atlas. Brown describes the role of CAD in clinical trials and the requirements for the CAD system to be used in a successful clinical trial. If a CAD detection system is to become clinically essential in the application to radiation treatment planning, it should be validated in oncologic clinical trials. However, CAD software alone is not sufficient for conducting a clinical trial. There are a number of architectural requirements such as receiving images (from multiple field sites), a fault-tolerance database for storing quantitative measures, and data-mining and reporting capabilities. Li discusses methods for improvement of bias and generalizability of CAD schemes. The reliable evaluation of CAD schemes is as important as the development of such schemes in the field of CAD research. In the past, the performance of CAD schemes has been evaluated by use of various methods such as resubstitution, leave-one-out, cross-validation, and hold-out. This article addresses important issues related to the bias and generalization performance of CAD schemes trained with limited datasets, and analysis of pitfalls in the incorrect use of various methods and approaches to the reduction of bias and variance.
Editorial / Computerized Medical Imaging and Graphics 31 (2007) 195–197
6. Security and integration The integration of CAD and PACS requires consideration of two issues: the method of integration and data security. Two papers by Zhou in this category discuss these topics: • Zhou et al.—CAD-PACS Integration Tool Kit Based on DICOM Secondary Capture, Structured Report and IHE Workflow Profiles; • Zhou—Data Security Assurance in CAD-PACS Integration. In his first paper, Zhou presents a self-contained DICOMbased toolkit based on the screen capture and the structured report of the DICOM standard; and IHE (Integrating Healthcare Enterprise) workflow profiles. His second paper provides a method to insure the integrity of image and CAD output data in the PACS environment by use of a lossless digital signatureembedding method. In summary, in addition to the editorial, the 15 papers included in this special issue on computer-aided diagnosis (CAD) and image-guided decision support covers six categories of state-of-the-art CAD research and development. It ranges
197
from a CAD overview and trends, targeted CAD applications, knowledge discovery and decision support, CAD systems and evaluation, to security and system integration. It is our belief that the integration of CAD with PACS will lead to the emergence of CAD applications into mainstream image-aided clinical practice in both diagnosis and treatment. Kunio Doi ∗ Kurt Rossmann Laboratories for Radiologic Image Research, Department of Radiology, The University of Chicago, 5841 South Maryland Avenue, Chicago, IL 60637, United States H.K. Huang Division of Imaging Informatics, Departments of Radiology and Biomedical Engineering, University of Southern California, 4676 Admiralty Way, Suite 601, Marina del Ray, CA 90292, United States ∗ Corresponding
author. Tel.: +1 773 702 6954; fax: +1 773 702 0371. E-mail addresses: [email protected] (K. Doi), [email protected] (H.K. Huang)