Integrated Diagnostics: The Computational Revolution Catalyzing Cross-disciplinary Practices in Radiology, Pathology, and Genomics
2017; Radiological Society of North America; Volume: 285; Issue: 1 Linguagem: Inglês
10.1148/radiol.2017170062
ISSN1527-1315
AutoresClaes Lundström, Hannah Gilmore, Pablo R. Ros,
Tópico(s)Artificial Intelligence in Healthcare and Education
ResumoHomeRadiologyVol. 285, No. 1 PreviousNext Reviews and CommentaryFree AccessOpinionIntegrated Diagnostics: The Computational Revolution Catalyzing Cross-disciplinary Practices in Radiology, Pathology, and GenomicsClaes F. Lundström , Hannah L. Gilmore, Pablo R. RosClaes F. Lundström , Hannah L. Gilmore, Pablo R. RosAuthor AffiliationsFrom the Center for Medical Image Science and Visualization, Linköping University Hospital, 581 85 Linköping, Sweden (C.F.L.); Sectra, Linköping, Sweden (C.F.L.); and Departments of Pathology (H.L.G.) and Radiology (P.R.R.), University Hospitals Cleveland Medical Center, Case Western Reserve University, Cleveland, Ohio.Address correspondence to C.F.L. (e-mail: [email protected]).Claes F. Lundström Hannah L. GilmorePablo R. RosPublished Online:Sep 19 2017https://doi.org/10.1148/radiol.2017170062MoreSectionsPDF ToolsImage ViewerAdd to favoritesCiteTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinked In AbstractIntroductionThe silo metaphor is an adequate description of current practices in the diagnostic specialties. Radiology, pathology, and genomics constitute three diagnostic disciplines with similar characteristics in terms of complex exploratory pathways. Despite their closeness and mutual interdependence, touching points are today remarkably few, and the support for close collaboration is meager. In recent years, however, voices have been raised calling for tighter collaboration creating deeply integrated workflows between radiology, pathology, and genomics. This multidisciplinary convergence is captured by the term integrated diagnostics (ID).We fully agree with the potential benefits of a development into ID practices that are characterized by profound teamwork on a daily basis, and in this article, we will describe the rationale for our standpoint. We focus primarily on the interplay of radiology and pathology and map out a viable and desirable scope for integrated practices across these two areas. We explore as well how genetics docks into the ID concepts.The concept of imaging-pathologic correlation is already close at heart for radiologists globally, who by training know the value of detailed correlation with pathologic findings. However, for many radiologists, this insight has no corollary in practical work, in the form of a rich and frequent exchange in clinical routine.This is the right time for a major move toward ID, because current technology can remove some hindrances and add new possibilities. One major change is that pathology diagnostics is transforming from an analog-slide-and-microscope approach to a digital workflow at a rapid pace thanks to whole-slide imaging scanners and pathology picture archiving and communication systems (PACS) for large-scale, clinical use. In Europe, several hospitals are already running digital workflows for routine primary review (1,2), and there are a vast amount of clinical digitization efforts ongoing across the globe. In the United States, many laboratories have begun to integrate digital pathology into practice, and there is a major effort underway to seek U.S. Food and Drug Administration approval for digital pathology as a primary diagnostic tool, as is done in Europe.With the use of digital pathology, the playing field for cross-disciplinary information technology (IT) tools greatly expands. Moreover, there is the strong trend for quantification of image contents to enable large-scale computational analysis. This is equally applicable for pathology and radiology, and in the latter case, it is known as radiomics (3). The disciplinary border actually becomes blurred and irrelevant when computational approaches such as deep learning are applied to quantified imaging features—the computational methods are the same regardless of the data source. In addition, the possibility of combining radiologic and pathologic imaging in machine learning approaches is a particularly promising aspect of ID.Three pillars for an effective ID practice are described in this article: steering of diagnostic pathways, concerted diagnostic decisions, and a unified interface to referring clinicians. The pillar descriptions are followed by a wish list for the development of supporting IT tools. The emphasis is not on blue-sky visions for the future, but rather on a practical, clinical agenda for change in disease diagnosis than can be materialized with existing technologies and organizations.Steering Diagnostic PathwaysThe pathway through diagnostics can be quite complex. Whereas standardized routes are desirable when possible, they fail to well represent many patient journeys. One aspect is that the patient may be referred to diagnostics several times in the course of finding the correct conclusion, in a fragmented process based on a referring specialist receiving partial results before ordering further examinations.The first pillar is to boost the role of the diagnostician in terms of pathway steering. In our design of an ID practice, the diagnosticians would take on a larger responsibility and would directly decide on follow-up studies when the next step in the pathway is clear. Whenever no mindset or organization barrier exists between disciplines, the most appropriate examination can be selected, regardless of whether the diagnostic examination is offered by radiology, pathology, or a laboratory-based test, including genomics. In this way, costs can potentially be reduced by shortening pathways and avoiding unnecessary studies. Overutilization is a well-known concern, and the need to guide referring clinicians is manifested within radiology by, for example, the appropriateness criteria of the American College of Radiology. An extended diagnostician role could be a complementary and very effective way to let the true experts implement desired diagnostic pathways. An ID approach to guidance would also avoid separating imaging from other diagnostic tests, which is a criticism voiced regarding the radiology criteria (4). Such a new paradigm poses a challenge, however, because referring physicians may fear that unnecessarily costly examination sets are chosen beyond their control. Perhaps this issue can be addressed by investing in close, trusted relationships, or perhaps payment models would need to be reformed, but nevertheless we argue that the empowered diagnostician role is worth pursuing.Krestin et al (5) argue that this role is suitable for the radiologist, initially for the benefit of the patient and ultimately for the benefit of radiology as a specialty. To that, we would add that teamwork among diagnosticians would be essential, because of the ever-increasing complexity of the diagnostic toolbox as it expands. The need is further accentuated as genomic testing continues to make its way into everyday diagnostic practice for an ever-increasing number of diseases. It has become clear that genomics does not mean a one-size-fits-all test but instead means carefully selected diagnostic panels guided by detailed findings in radiologic and pathologic imaging that must be put into the appropriate clinical context.The ID mindset also supports the coming flood of computational methods expected to be giant leaps forward for precision medicine. There is no reason to limit, for instance, predictive artificial intelligence methods, such as the very promising deep learning technology, to data from just one discipline. In our view, to run effectively computational examinations, a separate diagnostic specialty may be required. The role of this specialist would be very similar to that of other diagnosticians in general terms; the skill set would include selecting and designing appropriate studies and interpreting the results in the context of all other patient data. Special expertise is needed, however, to assess the benefit of computational methods in relation to the available data for the patient and comparable populations, and, above all, to understand the potential errors that can be made—in the same way as a radiologist knows about the potential for artifact creation in a computed tomography examination. In an ID department, such a role is natural, although it would be hampered if it were limited to just one of the silos. Jha and Topol (6) underline the importance of this future role of "information specialist." But while they envision a full merger of the radiology and pathology disciplines, including genomics, our view is that the need for medical specialization will not decrease even though the toolbox of computational diagnostic methods is expanded. Artificial intelligence will be a great new cross-disciplinary examination modality, but it will be a complement to, rather than a replacement for, the existing specialized diagnostic skill sets that require deep understanding of conventional examination methods.Finally, an important pathway challenge in everyday pathology is to determine whether a tissue sample is representative. When there is uncertainty about sampling error, the choice between potentially unnecessary additional biopsies or potentially missing disease is problematic. Sorace et al (7) highlight breast cancer and lung disease as particularly pertinent areas for joint radiology-pathology decision making. One example is interstitial lung disease: Tissue sampling error can be particularly misleading because different disease patterns can coexist in sampling sites, and some cases are indeterminate at histopathologic examination alone, even for specialized lung pathologists. Focusing on ID as a specific domain would pave the way for the development of novel IT solutions to highlight potential discordance where single-discipline systems fall short, which we discuss further in the next section.Concerted Diagnostic DecisionsThe second pillar of ID is facilitated joint radiology-pathology-genomics decision making. Today this is primarily taking place in multidisciplinary case conferences. We believe that ID can greatly enrich these conferences. When pathologic images are digitized, a basic demand is to have the images from both disciplines side by side, with clear and direct anatomic links between specimens and their correlations in the radiologic images. Likewise, with the increasing focus in oncology on genomic testing for optimal targeted therapy, showing the anatomic and morphologic context of these genomic changes will be highly beneficial for personalized medicine. New and improved IT tools can have a strong positive impact for the case conferences, boosting meeting efficiency at the same time as documenting vital components of the decision process. Computational methods can also assist with input on questions arising in the course of the discussion, such as eliciting actual outcomes for the institution's previous comparable patients for different treatment alternatives. In this way, multidisciplinary conferences such as tumor boards would become sessions where new insights are jointly developed, rather than sessions where predefined static presentations are shown.Another crucial aspect is discordance handling. Today, the communication between radiologist and pathologist is limited, and a discordance between their conclusions is not systematically handled. This is potentially hazardous for the quality of care. In addition, it is a missed opportunity for continuous learning on both sides that could reduce the risk of future misjudgments and inefficient diagnostic pathway decisions.Arnold et al (8) describe an innovative discordance-handling method based on auxiliary software and workflow that mandates that the radiologist deal with discordances between the disciplines. A desirable evolution of their concept would be to introduce this process broadly in routine workflows and to extend existing radiology information system (RIS)/PACS systems with supporting tools instead of adding a separate application.There are clear benefits for the pathologist in having easy access to radiology images when performing the diagnostic review, to efficiently correlate the specimen samples with their anatomic origins. Likewise, there would be clear benefits for the genomics discipline to have accessible imaging from pathology and radiology for correlative purposes and to guide further testing. For example, if the genomic alterations in a breast tumor suggest an indolent course without a need for chemotherapy, but the histopathologic and radiologic features are more characteristic of a high-grade tumor requiring aggressive therapy, there is a major clinical disconnect. Was there normal tissue contamination in the sample used for genomic analysis? Is there tumor hetereogeneity? Knowledge of and access to available genomic, radiologic, and pathologic features can enable physicians to identify and address discordant findings early to better personalize treatment.The increased interaction between radiologists and pathologists is in line with proposals for increased radiology outreach in general. Gunderman and Chou (9) describe four modes of radiologists' interaction, promoting a push toward actively seeking exchange with other clinicians. We echo this point of view and add pathology to the list of disciplines from which radiologists should seek reciprocally consultative relationships in clinical routine. In a recent study by Dickerson et al (10), direct in-person communication between radiologists and surgeons led to changed patient treatment in 43% of cases, a finding the authors mainly attribute to arriving at shared mental models. We should see similar positive effects with tighter teamwork between radiologists and pathologists.Unified Interface to Referring CliniciansThe long-term prosperity of radiology and pathology is dependent on the service level provided and perceived by the customers—the referring clinicians, patients, and, ultimately, society. The third ID pillar is a unified interface for all diagnostics toward the referrers and patients. In many other areas in society, a concierge role has been introduced as a main contact to guide customers through various service pathway organizations. This is a promising option for ID as well, in the form of an educated diagnostician who can discuss pathway options, inform on case progress, and deliberate on reports.Another component well worth exploring is providing a joint diagnostic report. The referring clinician wants a condensed and actionable response from all of the diagnostic modalities that will enable him or her make treatment decisions. Even if the radiologist and pathologist generate their respective report components, these could be merged and presented jointly. Genomics results would also be part of such a report, properly related to the imaging context and histopathologic findings. This could be particularly useful when overviewing prior examinations, where the result is much more relevant than the partial steps taken to arrive at that conclusion.The integrated diagnosis report should be tailored for integration with computational decision-support tools for the referring physician, an emerging realm of solutions (11) combining diagnostics information with both other patient information and the latest outcome statistics to help the physician make the optimal treatment decision. Such decision support would also be well suited for tumor board guidance together with the diagnosticians.Technology Development AgendaThe ID scope outlined entails a set of required IT tools, some of which are yet to be developed. Because additional IT applications are problematic owing to increasing infrastructure complexity, we argue that the functions should primarily be added to existing RIS/PACS platforms and closely integrated with electronic health record systems. The main points of the development agenda are as follows:1. Integrated access to images and reports of all disciplines, using a single interface.2. Machine learning handling cross-disciplinary data, including imaging as well as genomics data, for diagnostic assistance.3. Ability to connect and follow findings, such as specific lesions, across disciplines.4. Correlation of imaging features, such as lesion size, shape, and structure, across disciplines.5. Cross-disciplinary discordance presentation and resolution.6. Automated merging of radiologic, pathologic, and genomic findings into a joint report.7. Support for decision process documentation at tumor boards.ConclusionThe opportunities described above sum up to a strong call for focusing on ID as one of the key forefronts of health care development. The obstacles are not technology limitations, because the agenda outlined is well within grasp with state-of-the-art methods, but rather in traditional approaches and inflexible organizational models.Thus, innovative institutions deploying effective change management should immediately be able to start reaping benefits.Finally, let us not forget the third major diagnostic component. Genomics should benefit by following the same path toward a deeper integration of genotype and phenotype, in both radiologic and histologic terms, resulting in the highest possible quality of health care.Disclosures of Conflicts of Interest: C.F.L. Activities related to the present article: is an employee of Sectra. Activities not related to the present article: holds stock in Sectra. Other relationships: disclosed no relevant relationships. H.L.G. disclosed no relevant relationships. P.R.R. Activities related to the present article: disclosed no relevant relationships. Activities not related to the present article: has received research fellowship support to University Hospitals Cleveland Medical Center/Case Western Reserve University from Sectra. Other relationships: disclosed no relevant relationships.References1. Thorstenson S, Molin J, Lundström C. Implementation of large-scale routine diagnostics using whole slide imaging in Sweden: digital pathology experiences 2006-2013. J Pathol Inform 2014;5(1):14. Crossref, Medline, Google Scholar2. Stathonikos N, Veta M, Huisman A, van Diest PJ. Going fully digital: perspective of a Dutch academic pathology lab. J Pathol Inform 2013;4:15. Crossref, Medline, Google Scholar3. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology 2016;278(2):563–577. Link, Google Scholar4. Jamal T, Gunderman RB. The American College of Radiology Appropriateness Criteria: the users' perspective. J Am Coll Radiol 2008;5(3):158–160. Crossref, Medline, Google Scholar5. Krestin GP, Grenier PA, Hricak H, et al. Integrated diagnostics: proceedings from the 9th biennial symposium of the International Society for Strategic Studies in Radiology. Eur Radiol 2012;22(11):2283–2294. Crossref, Medline, Google Scholar6. Jha S, Topol EJ. Adapting to artificial intelligence: radiologists and pathologists as information specialists. JAMA 2016;316(22):2353–2354. Crossref, Medline, Google Scholar7. Sorace J, Aberle DR, Elimam D, Lawvere S, Tawfik O, Wallace WD. Integrating pathology and radiology disciplines: an emerging opportunity? BMC Med 2012;10(1):100. Crossref, Medline, Google Scholar8. Arnold CW, Wallace WD, Chen S, et al. RadPath: a web-based system for integrating and correlating radiology and pathology findings during cancer diagnosis. Acad Radiol 2016;23(1):90–100. Crossref, Medline, Google Scholar9. Gunderman RB, Chou HY. The future of radiology consultation. Radiology 2016;281(1):6–9. Link, Google Scholar10. Dickerson EC, Alam HB, Brown RK, Stojanovska J; Michigan Radiology Quality Collaborative, Davenport MS. In-person communication between radiologists and acute care surgeons leads to significant alterations in surgical decision making. J Am Coll Radiol 2016;13(8):943–949. Crossref, Medline, Google Scholar11. Castaneda C, Nalley K, Mannion C, et al. Clinical decision support systems for improving diagnostic accuracy and achieving precision medicine. J Clin Bioinforma 2015;5:4. Crossref, Medline, Google ScholarArticle HistoryReceived January 9, 2017; revision requested February 20; revision received March 16; accepted April 4; final version accepted April 9.Published online: Sept 19 2017Published in print: Oct 2017 FiguresReferencesRelatedDetailsCited ByFuture of Business and FinanceTim-RasmusKiehl20225G IoT and Edge Computing for Smart HealthcareAbdulhamitSubasi, Siba SmarakPanigrahi, Bhalchandra SunilPatil, M. AbdullahCanbaz, RikuKlén2022Volumetric Tissue Imaging of Surgical Tissue Specimens Using Micro–Computed TomographyAndreas SPapazoglou, EfstratiosKaragiannidis, AlexandrosLiatsos, AndreanaBompoti, Dimitrios VMoysidis, ChristosArvanitidis, FaniTsolaki, SokratisTsagkaropoulos, StamatiosTheocharis, GeorgiosTagarakis, James SMichaelson, Markus DHerrmann2022 | American Journal of Clinical PathologyState-of-the-art review of lung imaging in cystic fibrosis with recommendations for pulmonologists and radiologists from the "iMAging managEment of cySTic fibROsis" (MAESTRO) consortiumPierluigiCiet, SilviaBertolo, MircoRos, RosariaCasciaro, MarcoCipolli, StefanoColagrande, StefanoCosta, ValeriaGalici, AndreaGramegna, CeciliaLanza, FrancescaLucca, LetiziaMacconi, FabioMajo, AntonellaPaciaroni, Giuseppe FabioParisi, FrancescaRizzo, IgnazioSalamone, TeresaSantangelo, LuigiaScudeller, LucaSaba, PaoloTomà, GiovanniMorana2022 | European Respiratory Review, Vol. 31, No. 163An integrative ultrasound-pathology approach to improve preoperative phyllodes tumor classification: A pilot studyPaolaLocicero, NoëlleWeingertner, VincentNoblet, MarieMondino, CaroleMathelin, SébastienMolière2022 | Breast Disease, Vol. 41, No. 1Lecture Notes in Networks and SystemsMadhuriThimmapuram, SowjanyaPentakota, H.Naga Chandrika2021 | , Vol. 210Baseline PET/CT imaging parameters for prediction of treatment outcome in Hodgkin and diffuse large B cell lymphoma: a systematic reviewR.Frood, C.Burton, C.Tsoumpas, A. F.Frangi, F.Gleeson, C.Patel, A.Scarsbrook2021 | European Journal of Nuclear Medicine and Molecular Imaging, Vol. 48, No. 10Die Mikroarchitektur des Pankreaskarzinoms aus Sicht des Pathologen und des RadiologenPhilippMayer, Matthias M.Gaida2021 | Der Pathologe, Vol. 42, No. 5Staging and Classification of Primary Musculoskeletal Bone and Soft-Tissue Tumors According to the 2020 WHO Update, From the AJR Special Series on Cancer StagingMark DouglasMurphey, Mark JayKransdorf2021 | American Journal of Roentgenology, Vol. 217, No. 5The Enterprise Imaging Value PropositionCheryl A.Petersilge2020 | Journal of Digital Imaging, Vol. 33, No. 1An Electronic Form for Reporting Results of Targeted Prostate Biopsy: Urology Integrated Diagnostic Report (Uro-IDR)JacobGuorgui, AdamKinnaird, RajivJayadevan, Alan M.Priester, Corey W.Arnold, Leonard S.Marks2020 | Urology, Vol. 138Integrated diagnosticsGiuseppeLippi, MarioPlebani2020 | Biochemia medica, Vol. 30, No. 1Segmentation and Classification in Digital Pathology for Glioma Research: Challenges and Deep Learning ApproachesTahsinKurc, SpyridonBakas, XuhuaRen, AdityaBagari, AlexandreMomeni, YueHuang, LichiZhang, AshishKumar, MarcThibault, QiQi, QianWang, AvinashKori, OlivierGevaert, YunlongZhang, DinggangShen, MahendraKhened, XinghaoDing, GanapathyKrishnamurthi, JayashreeKalpathy-Cramer, JamesDavis, TianhaoZhao, RajarsiGupta, JoelSaltz, KeyvanFarahani2020 | Frontiers in Neuroscience, Vol. 14Radiología y patología: de viejas amigas a aliadas estratégicasAurelioAriza2019 | Radiología, Vol. 61, No. 2Radiology and pathology: From old friends to strategic partnersA.Ariza2019 | Radiología (English Edition), Vol. 61, No. 2Rise of the Machines: Advances in Deep Learning for Cancer DiagnosisAdrian B.Levine, ColinSchlosser, JasleenGrewal, RobinCoope, Steve J.M.Jones, StephenYip2019 | Trends in Cancer, Vol. 5, No. 3Lessons learnt from pathologic imaging correlation in the liver: an historical perspectiveYvonnePurcell, PaulineCopin, ValérieParadis, ValérieVilgrain, MaximeRonot2019 | The British Journal of Radiology, Vol. 92, No. 1097The future of radiology augmented with Artificial Intelligence: A strategy for successCharleneLiew2018 | European Journal of Radiology, Vol. 102Deep learning for lung cancer prognostication: A retrospective multi-cohort radiomics studyAhmedHosny, ChintanParmar, Thibaud P.Coroller, PatrickGrossmann, RomanZeleznik, AvnishKumar, JohanBussink, Robert J.Gillies, Raymond H.Mak, Hugo J. W. L.Aerts, Atul J.Butte2018 | PLOS Medicine, Vol. 15, No. 11Implementing the DICOM Standard for Digital PathologyMarkus D.Herrmann, David A.Clunie, AndriyFedorov, Sean W.Doyle, StevenPieper, VeronicaKlepeis, Long PLe, George L.Mutter, David S.Milstone, Thomas J.Schultz, RonKikinis, Gopal K.Kotecha, David H.Hwang, Katherine PAndriole, A. Johnlafrate, James A.Brink, Giles W.Boland, Keith J.Dreyer, MarkMichalski, Jeffrey A.Golden, David N.Louis, Jochen K.Lennerz2018 | Journal of Pathology Informatics, Vol. 9, No. 1Recommended Articles Integrating Al Algorithms into the Clinical WorkflowRadiology: Artificial Intelligence2021Volume: 3Issue: 6Customer Service in Radiology: Satisfying Your Patients and ReferrersRadioGraphics2018Volume: 38Issue: 6pp. 1872-1887Multimedia-enhanced Radiology Reports: Concept, Components, and ChallengesRadioGraphics2018Volume: 38Issue: 2pp. 462-482Implementing a Radiology Residency Quality Curriculum to Develop Physician Leaders and Increase Value for PatientsRadioGraphics2020Volume: 40Issue: 2pp. 505-514Deep Learning: A Primer for RadiologistsRadioGraphics2017Volume: 37Issue: 7pp. 2113-2131See More RSNA Education Exhibits Keys to Starting LI-RADS in Your Practice: Tips from the Society of Abdominal Radiology-HCC Diagnosis Disease Focused Panel Digital Posters2020Applications Of Domain-Inspired Radiomics And Deep Learning In NeuroradiologyDigital Posters2021Artificial Intelligence in Radiology: A Primer for Residents and StudentsDigital Posters2019 RSNA Case Collection Giant cell tumor RSNA Case Collection2020Diffuse idiopathic pulmonary neuroendocrine cell hyperplasiaRSNA Case Collection2020Diffuse idiopathic pulmonary neuroendocrine cell hyperplasiaRSNA Case Collection2020 Vol. 285, No. 1 Funding/SupportC.F.L. supported by VINNOVA (2014-04257). AbbreviationsAbbreviations:IDintegrated diagnosticsITinformation technologyPACSpicture archiving and communication systemRISradiology information system Metrics Altmetric Score PDF download
Referência(s)