Skip to main content
SpringerLink
Log in

Digitalisierte Bildverarbeitung: künstliche Intelligenz im diagnostischen Einsatz

Digitalized image processing: artificial intelligence in diagnostic procedures

  • Fokus
  • Published:
Forum Aims and scope

Zusammenfassung

Die digitalisierte Bildverarbeitung hat in den vergangenen Jahren große Fortschritte erzielt. Erste Systeme zur automatisierten Bildanalyse basierten auf mehrstufigen Algorithmen des Maschinenlernens. Diese wurden inzwischen zu selbstlernenden neuronalen Netzwerken weiterentwickelt. Solche auf künstlicher Intelligenz (KI) basierenden Systeme werden zunehmend auch in der Medizin eingesetzt. Die Dermatologie bietet sich durch die einfache Zugänglichkeit der Hautoberfläche für eine KI-basierte Diagnosestellung durch Bildanalyse an. Innerhalb der Dermatologie nimmt die Hautkrebsdiagnostik einen zentralen Stellenwert ein. Dermatoskopische Aufnahmen von pigmentierten und nichtpigmentierten Hautveränderungen sind in großen öffentlichen Datenbanken verfügbar und können so zum Training automatisierter Diagnosesysteme eingesetzt werden. In Studien zur Hautkrebsdiagnostik erzielten verschiedene KI-basierte Systeme eine diagnostische Genauigkeit, welche mit der von erfahrenen Dermatologen vergleichbar war. Ein erstes neuronales Netzwerk zur Hautkrebsdiagnostik anhand von dermatoskopischen Bildern ist zwischenzeitlich für den europäischen Markt zugelassen worden (Moleanalyzer pro®, FotoFinder Systems GmbH). Auch in der Dermatohistopathologie wurden KI-basierte Diagnosesysteme entwickelt und in Studien getestet. Trotz aller Euphorie über die hohe Leistungsfähigkeit ist es jedoch von essenzieller Bedeutung, auch Limitierungen KI-basierter Diagnosesysteme aufzuzeigen. So beeinflussen Bildartefakte wie farbliche Markierungen oder eingeblendete Lineale innerhalb dermatoskopischer Aufnahmen die Bildbeurteilung durch neuronale Netzwerke. Genaue Kenntnisse über die Limitierungen der neuen Technologie sind entscheidend, um Indikationen und Kontraindikationen von KI-basierten Diagnosesystemen zu bestimmen und die Leistungsfähigkeit weiter zu verbessern. Auch wenn neuronale Netzwerke bereits als Assistenzsysteme für die Hautkrebserkennung eingesetzt werden, liegen der verantwortungsvolle Umgang mit den Ergebnissen und die abschließende Therapieentscheidung weiterhin beim klinisch tätigen Arzt.

Abstract

Digital image processing has made much progress during recent years. First systems for automated image analysis were based on multistage machine learning algorithms and required extensive preprocessing and segmentation using handcrafted filters. Meanwhile, current systems are based on deep-learning convolutional neuronal networks (CNNs). Such artificial intelligence (AI) based systems are increasingly being established for use in clinical medicine. Dermatology is especially suitable for digital image analysis since human skin can easily be accessed. Due to continuously increasing incidence rates the diagnosis of skin cancer is of special interest in dermatology. Dermoscopic images of pigmented and non-pigmented skin lesions have been collected in large publicly available databases and may be used for training of automated diagnostic systems. Previous clinical studies have shown that CNNs are capable of attaining a dermatologist level diagnostic accuracy in the detection of skin cancer. A first CNN has recently been approved as a medical device for the European market (Moleanalyzer pro®, FotoFinder Systems GmbH). Moreover, deep-learning CNNs have also been trained and tested in the field of dermatohistopathology. Despite the euphoria about the high-level diagnostic performance, it is crucial to carefully evaluate the limitations of AI-based diagnostic systems. Image artefacts, such as color markings and superimposed scales that are included in the image may severely impair the diagnostic accuracy. Investigating such limitations is relevant to determine the indications and contraindications and to further improve the performance of AI-based diagnostic systems. Importantly, AI-based diagnostic devices are intended as assistance systems and the results should be carefully integrated into the decision-making progress by clinicians, who remain fully responsible for the management decisions.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Abb. 1
Abb. 2
Abb. 3
Abb. 4

Literatur

  1. Apalla Z et al (2017) Epidemiological trends in skin cancer. Dermatol Pract Concept 7(2):1–6

    Article  Google Scholar 

  2. Stolz W (1994) ABCD rule of dermatoscopy: a new practical method for early recognition of malignant melanoma. Eur J Dermatol 4:521–527

    Google Scholar 

  3. Gachon J et al (2005) First prospective study of the recognition process of melanoma in dermatological practice. Arch Dermatol 141(4):434–438

    Article  Google Scholar 

  4. Pehamberger H, Steiner A, Wolff K (1987) In vivo epiluminescence microscopy of pigmented skin lesions. I. Pattern analysis of pigmented skin lesions. J Am Acad Dermatol 17(4):571–583

    Article  CAS  Google Scholar 

  5. Menzies SW et al (1996) Frequency and morphologic characteristics of invasive melanomas lacking specific surface microscopic features. Arch Dermatol 132(10):1178–1182

    Article  CAS  Google Scholar 

  6. Argenziano G et al (2011) Seven-point checklist of dermoscopy revisited. Br J Dermatol 164(4):785–790

    Article  CAS  Google Scholar 

  7. Dick V et al (2018) Bildbasierte Computerdiagnose des Melanoms. Hautarzt 69(7):591–601

    Article  CAS  Google Scholar 

  8. Pathan S, Prabhu KG, Siddalingaswamy P (2018) Techniques and algorithms for computer aided diagnosis of pigmented skin lesions—a review. Biomed Signal Process Control 39:237–262

    Article  Google Scholar 

  9. Fujisawa Y, Inoue S, Nakamura Y (2019) The possibility of deep learning-based, computer-aided skin tumor classifiers. Front Med 6:191

    Article  Google Scholar 

  10. Fujisawa Y et al (2019) Deep-learning-based, computer-aided classifier developed with a small dataset of clinical images surpasses board-certified dermatologists in skin tumour diagnosis. Br J Dermatol 180(2):373–381

    Article  CAS  Google Scholar 

  11. Baltruschat IM et al (2019) Comparison of deep learning approaches for multi-label chest X‑ray classification. Sci Rep 9(1):6381

    Article  Google Scholar 

  12. Nasr-Esfahani E et al (2016) Melanoma detection by analysis of clinical images using convolutional neural network. Conf Proc IEEE Eng Med Biol Soc. 2016:1373-1376. https://doi.org/10.1109/EMBC.2016.7590963

    Article  CAS  PubMed  Google Scholar 

  13. Arribas JI, Cid-Sueiro J (2005) A model selection algorithm for a posteriori probability estimation with neural networks. IEEE Trans Neural Netw 16(4):799–809

    Article  Google Scholar 

  14. Tschandl P et al (2019) Diagnostic accuracy of content-based dermatoscopic image retrieval with deep classification features. Br J Dermatol 181(1):155–165. https://doi.org/10.1111/bjd.17189. Epub 2018 Oct 17

    Article  CAS  Google Scholar 

  15. Handelman G et al (2018) eD octor: machine learning and the future of medicine. J Intern Med 284(6):603–619

    Article  CAS  Google Scholar 

  16. Weiss K, Khoshgoftaar TM, Wang D (2016) A survey of transfer learning. J Big Data 3(1):9

    Article  Google Scholar 

  17. Tschandl P, Rosendahl C, Kittler H (2018) The HAM10000 Dataset: a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Sci Data. 5:180161. https://doi.org/10.1038/sdata.2018.161

    Article  PubMed  PubMed Central  Google Scholar 

  18. Han SS et al (2018) Classification of the clinical images for benign and malignant cutaneous tumors using a deep learning algorithm. J Invest Dermatol 138(7):1529–1538

    Article  CAS  Google Scholar 

  19. Esteva A et al (2017) Dermatologist-level classification of skin cancer with deep neural networks. Nature 542(7639):115–118

    Article  CAS  Google Scholar 

  20. Haenssle H et al (2018) Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. Ann Oncol 29(8):1836–1842

    Article  CAS  Google Scholar 

  21. Brinker TJ et al (2019) A convolutional neural network trained with dermoscopic images performed on par with 145 dermatologists in a clinical melanoma image classification task. Eur J Cancer 111:148–154

    Article  Google Scholar 

  22. Tschandl P et al (2019) Comparison of the accuracy of human readers versus machine-learning algorithms for pigmented skin lesion classification: an open, web-based, international, diagnostic study. Lancet Oncol 20(7):938–947. https://doi.org/10.1016/S1470-2045(19)30333-X. Epub 2019 Jun 12

    Article  Google Scholar 

  23. Tschandl P et al (2019) Expert-level diagnosis of nonpigmented skin cancer by combined convolutional neural networks. JAMA Dermatol 155(1):58–65

    Article  Google Scholar 

  24. Gertych A et al (2019) Convolutional neural networks can accurately distinguish four histologic growth patterns of lung adenocarcinoma in digital slides. Sci Rep 9(1):1483

    Article  Google Scholar 

  25. Hekler A et al (2019) Pathologist-level classification of histopathological melanoma images with deep neural networks. Eur J Cancer 115:79–83

    Article  Google Scholar 

  26. Kulkarni PM et al (2019) Deep learning based on standard H&E images of primary melanoma tumors identifies patients at risk for visceral recurrence and death. Clin Cancer Res. https://doi.org/10.1158/1078-0432.CCR-19-1495. [Epub ahead of print]

    Article  PubMed  Google Scholar 

  27. Sun M et al (2019) Prediction of BAP1 expression in uveal melanoma using densely-connected deep classification networks. Cancers 11(10):1579

    Article  Google Scholar 

  28. Gavrilov AD et al (2018) Preventing model overfitting and underfitting in convolutional neural networks. Int J Softw Sci Comput Intell 10(4):19–28

    Article  Google Scholar 

  29. Cogswell M et al (2015) Reducing overfitting in deep networks by decorrelating representations. https://arxiv.org/abs/1511.06068

  30. Winkler JK et al (2019) Association between surgical skin markings in dermoscopic images and diagnostic performance of a deep learning convolutional neural network for melanoma recognition. JAMA Dermatol. https://doi.org/10.1001/jamadermatol.2019.1735. [Epub ahead of print]

    Article  PubMed  Google Scholar 

  31. Narla A et al (2018) Automated classification of skin lesions: from pixels to practice. J Invest Dermatol 138(10):2108–2110

    Article  CAS  Google Scholar 

  32. Hekler A et al (2019) Superior skin cancer classification by the combination of human and artificial intelligence. Eur J Cancer 120:114–121

    Article  Google Scholar 

  33. Fink C et al (2018) Patient acceptance and trust in automated computer-assisted diagnosis of melanoma with dermatofluoroscopy. J Dtsch Dermatol Ges 16(7):854–859

    PubMed  Google Scholar 

Download references

Author information

Authors and Affiliations

  1. Universitätshautklinik Heidelberg, Im Neuenheimer Feld 440, 69120, Heidelberg, Deutschland

    J. K. Winkler, K. Sies, C. Fink, F. Toberer, A. Enk &  H. A. Haenssle

Authors
  1. J. K. Winkler
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. K. Sies
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. C. Fink
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. F. Toberer
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. A. Enk
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. H. A. Haenssle
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to H. A. Haenssle.

Ethics declarations

Interessenkonflikt

H. A. Haenssle erhielt Honorare und/oder Reisekostenunterstützung von Firmen, die an der Entwicklung von Systemen zum Hautkrebsscreening beteiligt sind: Scibase AB, FotoFinder Systems GmbH, Heine Optotechnik GmbH, Magnosco GmbH. C. Fink erhielt Reisekostenunterstützung von Magnosco GmbH. J. K. Winkler, K. Sies, F. Toberer und A. Enk geben an, dass kein Interessenkonflikt besteht.

Alle beschriebenen Untersuchungen am Menschen oder an menschlichem Gewebe wurden mit Zustimmung der zuständigen Ethikkommission, im Einklang mit nationalem Recht sowie gemäß der Deklaration von Helsinki von 1975 (in der aktuellen, überarbeiteten Fassung) durchgeführt. Von allen beteiligten Patienten liegt eine Einverständniserklärung vor.

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Winkler, J.K., Sies, K., Fink, C. et al. Digitalisierte Bildverarbeitung: künstliche Intelligenz im diagnostischen Einsatz. Forum 35, 109–116 (2020). https://doi.org/10.1007/s12312-019-00729-3

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s12312-019-00729-3

Schlüsselwörter

Keywords

Search

Navigation