Machine Learning Applications in Healthcare


Bayram M. S. B., Akın A.

24th Biomedical Science and Technology Symposium (BIOMED2019), İzmir, Türkiye, 17 - 20 Ekim 2019, ss.1

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: İzmir
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.1
  • Acıbadem Mehmet Ali Aydınlar Üniversitesi Adresli: Evet

Özet

INTRODUCTION

The use of Clinical Decision Support Systems in healthcare is leading the rapidly increasing applications. According to a report published by the US research company The CB Insights in July 20191, a total of 66 billion US dollars of funding has been provided for the start of approximately 3600 machine learning (ML) startups in the 6-year period from the second quarter of 2013 to the second quarter of 2019. This trend has been going on for the last 6 years and supported startups are developing products that use artificial intelligence especially for diagnosis, drug R&D, remote patient monitoring, emergency room and hospital management, genomic and patient risk scoring.

Here, we present some of our preliminary findings based on ML and Deep Learning studies in healthcare; especially in dermatology, radiology and ophthalmology, also natural language processing to make patient-doctor communication more continuous and more efficient. Predicting athlete injuries and athlete performance in a professional football team will be discussed.

MATERIALS & METHODS

One of the studies we have been exploring was using the open source isic-archive dataset2, dermoscopy images of nevi or malenoma classified as benign or malignant based on the images which have been verified by the histopathology results. A total of 4304 images were evalualted (2436 nevi, 1868 melanoma). The study of ML-based prediction was whether a skin lesion is nevi or melanoma and did not predict any subclass of the lesion itself.

RESULTS AND DISCUSSION

The comparison of two deep neural network algorithms using convolutional neural networks suggested that, both have their strengths in different areas. MobileNet is more efficient to even run on slower computers with limited resources and had better accuracy while Inception had better sensitivity suggesting in the early

stages of diagnostic processes, it could work  more reliably in a clinical decision support system. 

CONCLUSION

Our results were in line with the literature that the digital imaging based healthcare data could benefit the most from ML-based clinical decision support systems. For ML applications, the data is everything, the more the better. Therefore it is crucial to having access more data that would definitely increase the efficiency of the algorithms and accuracy of the results.