Machine Learning

We have studied and applied various machine learning methods (e.g., CBR,SVM/SVR) in a number of clinical contexts, with one focus on time series analysis. More recently we have studied deep and convolutional neural networks for classification and regression tasks based on sensor and image data. Particularly, we extend deep learning methods to spatio-temporal data, including 4D image data. Application examples range from nephrology to tumor tissue detection to cardiology.

Selected publications

N. Gessert, M. Schlüter, A. Schlaefer (2018). A Deep Learning Approach for Pose Estimation from Volumetric OCT Data. Medical Image Analysis. 46 162-179.

N. Gessert, T. Priegnitz, T. Saathoff, S.-T. Antoni, D. Meyer, M. F. Hamann, K.-P. Jünemann, C. Otte, A. Schlaefer (2018). Needle Tip Force Estimation using an OCT Fiber and a Fused convGRU-CNN Architecture - MICCAI 2018. International Conference on Medical Image Computing and Computer-Assisted Intervention 222-229, Spotlight Talk.

C. Otte, S. Otte, L. Wittig, G. Hüttmann, C. Kugler, D. Drömann, A. Zell, A. Schlaefer (2014). Investigating Recurrent Neural Networks for OCT A-scan Based Tissue Analysis. Methods Inf Med. 53 (4), 245-249.

A. Schlaefer, S. Dieterich (2011). Feasibility of case-based beam generation for robotic radiosurgery. Artif Intell Med. 52 (2), 67-75.

L. Fritsche, A. Schlaefer, K. Budde, K. Schroeter, H. H. Neumayer (2002). Recognition of critical situations from time series of laboratory results by case-based reasoning. J Am Med Inform Assoc. 9 (5), 520-528.

Tissue Characterization

Mainly, medical decisions rely on histological analysis of the tissue. The correct analysis depends on the provided sample of tissue which needs to be classified. Retrieving reliable tissue samples, e.g., by biopsy, is a challenging task. Hence, we study methods for biopsy guidance, i.e., using in-needle imaging techniques like optical coherence tomography.

Tissue Classification using Optical Coherence Tomography

TC_OCT_NN.png

Linear mapping of OCT A-scans along the insertion axis. The color coding is based on the classification result of a trained neural network. Red indicates A-scans with a high tumor probability. The white marker indicated mm and cm, respectively.

The needle biopsy is an established and sparing technique. But with regard to applications, e.g. the transbronchial needle biopsy (TBNA) only has a 70% chance to confirm a diagnosis. Especially regarding small peripheral nodules, the placement of biopsy needles is challenging. Optical Coherence Tomography (OCT) has been proposed to support biopsy guidance. OCT has a high spatial and temporal resolution and can be implemented in needle tips through single optical fibers.

We study how OCT can be used to classify tissue in order to improve diagnostics or guidance in biopsy. We developed a neural network algorithm to classify tissue with high tumor probability solely with the help of OCT imaging.

Contact

Opens internal link in new windowAlexander Schlaefer (schlaefer(at)tuhh.de)

References

  • A. Patel, C. Otte, A. Schlaefer, D. Nir, S. Otte, T. Ngo, T. Loke, M. Winkler (2016). MP34-14 Investigating the feasibility of optical coherence tomography to identify prostate cancer - an ex-vivo study. Presented at the Annual Meetig of American Urological Association (AUA) [BibTex]

  • C. Otte, S. Otte, L. Wittig, G. Hüttmann, C. Kugler, D. Drömann, A. Zell, A. Schlaefer (2014). Investigating Recurrent Neural Networks for OCT A-scan Based Tissue Analysis. Methods Inf Med. 53 (4), [doi] [BibTex]

  • L. Wittig, C. Otte, S. Otte, G. Hüttmann, D. Drömann, A. Schlaefer (2014). Tissue analysis of solitary pulmonary nodules using OCT A-Scan imaging needle probe. The European respiratory journal. [www] [BibTex]

  • C. Otte, S. Otte, L. Wittig, G. Hüttmann, D. Drömann, A. Schlaefer (2013). Identifizierung von Tumorgewebe in der Lunge mittels optischer Kohärenztomographie. [BibTex]

  • S. Otte, C. Otte, A. Schlaefer, L. Wittig, G. Huttmann, D. Drömann, A. Zell (2013). OCT A-Scan based lung tumor tissue classification with Bidirectional Long Short Term Memory networks. Machine Learning for Signal Processing (MLSP), 2013 IEEE International Workshop on 1-6. [doi] [www] [BibTex]