Active Machine Learning of Complex Visual Tasks

This paper reports on the development of an artificial vision system implemented in software and its application to mammography. It describes a supervision strategy that facilitates the machinecentered learning of complex visual tasks. The key contributions of this paper are the description of our “active” learning strategy and a mechanism by which pixels associated with individual artifacts visible to a human eye in an image can be captured and used as training examples for a machinelearning algorithm. Techniques are discussed in the context of the analysis of micro-calcifications. The significance is that it provides a means by which illdefined concepts (e.g. visual characteristics of tumors) that are embedded in a complex image (e.g. mammograms) can be more efficiently and accurately learned by a machine.

[1]  F. Bray,et al.  The changing global patterns of female breast cancer incidence and mortality , 2004, Breast Cancer Research.

[2]  C. Baines,et al.  A tangled web: factors likely to affect the efficacy of screening mammography. , 1999, Journal of the National Cancer Institute.

[3]  Thomas G. Dietterich.  Multiple Classifier Systems , 2000, Lecture Notes in Computer Science.

[4]  Tom Hintz,et al.  Pseudo-invariant image transformations on a hexagonal lattice , 2000, Image Vis. Comput..

[5]  Eric L. Schwartz,et al.  Computational anatomy and functional architecture of striate cortex: A spatial mapping approach to perceptual coding , 1980, Vision Research.

[6]  L. Keith,et al.  Are mammography and palpation sufficient for breast cancer screening? A dissenting opinion. , 2002, Journal of women's health & gender-based medicine.

[7]  David R. Williams,et al.  Seeing through the photoreceptor mosaic , 1986, Trends in Neurosciences.

[8]  C. Mettlin,et al.  Global breast cancer mortality statistics , 1999, CA: a cancer journal for clinicians.

[9]  B. Zheng,et al.  Soft-copy mammographic readings with different computer-assisted detection cuing environments: preliminary findings. , 2001, Radiology.

[10]  C. D'Orsi,et al.  Accuracy of screening mammography interpretation by characteristics of radiologists. , 2004, Journal of the National Cancer Institute.

[11]  Benjamin B. Bederson,et al.  A miniaturized space-variant active vision system: Cortex-I , 1995, Machine Vision and Applications.

[12]  OpitzDavid,et al.  Popular ensemble methods , 1999 .

[13]  Ivars Peterson.  THE HONEYCOMB CONJECTURE , 1999 .

[14]  Thomas C. Hales,et al.  The Honeycomb Conjecture , 1999, Discret. Comput. Geom..

[15]  Andrew W. Moore,et al.  Reinforcement Learning: A Survey , 1996, J. Artif. Intell. Res..

[16]  Mark Last,et al.  Improving Stability of Decision Trees , 2002, Int. J. Pattern Recognit. Artif. Intell..

[17]  C J D'Orsi,et al.  Computer-aided detection: there is no free lunch. , 2001, Radiology.

[18]  John J. Grefenstette,et al.  Evolutionary Algorithms for Reinforcement Learning , 1999, J. Artif. Intell. Res..

[19]  E. Thurfjell,et al.  Sensitivity and specificity of computer-assisted breast cancer detection in mammography screening , 1998, Acta radiologica.

[20]  D. Miglioretti,et al.  Individual and Combined Effects of Age, Breast Density, and Hormone Replacement Therapy Use on the Accuracy of Screening Mammography , 2003, Annals of Internal Medicine.

[21]  J. Elmore,et al.  Accuracy of screening mammography using single versus independent double interpretation. , 2000, AJR. American journal of roentgenology.