Unsupervised Deep Learning Approach in Medical ImageAnalysis and Diagnosis

Abstract

Digital medical imaging is now one among the most important area and medical decision support systems has
an important role based on the size of the medical image. So detail analysis and deep learning in those images
are needed. Medical imaging data’s are often difficult to get since the common people are not ready to reveal
those data to the public. Treatment of complex disease after diagnosis from the medical image data is really a
key challenge in the development of health care sector. In the area of medical image diagnosis and analysis for
the past few years we have attained much promising results in deep learning methods. Better results are
obtained in both supervised and unsupervised deep learning technique. While comparing supervised with
unsupervised, unsupervised deep learning do not required manual efforts for creating separate class label based
on the algorithm and it derives detailed information from the input data which is given to the system itself.
Any of the external bias for decision making is not needed and it’s purely a data driven decision in case of an
unsupervised learning technique. In this paper wediscuss in detail about various types of unsupervised
methods for medical image analysis. Future challenges and researches in unsupervised deep learning
techniques of medical images are also being discussed in this paper. The best unsupervised learning method
can be determined based on the F1 score. For that scores of the unsupervised deep learning methods can be
compared with the popular computer vision methods used earlier. The major problems faced during medical
image information retrieval, semi supervised deep learning techniques and fusion of unsupervised and semi
supervised deep learning techniques and further steps to avoid the drawbacks of those problems can also be
discussed in detail in thispaper.