Breast Cancer Research


LESION DETECTION

Convolutional Neural Network (CNN) architecture used in the 1990s. 

The CNNs are the base structure of current deep learning architectures.


CNN for CADe of microcalcifications.

CNN for CADe of masses

CADView – our first CAD system for prospective clinical trial in screening mammography (1990s)

Digital Breast Tomosynthesis (DBT)

Prototype DBT System

LMLO View

LCC View

Computer-aided detection of breast mass in digital breast tomosynthesis (DBT) volume. 

A.  Pre-screening for lesion candidates

B.  Feature-based false-positive reduction

C.  DCNN-based false-positive reduction

Breast-based FROC curves for the feature-based CAD and DCNN-based CAD systems on the DBT test set (Ts). Each lesion in a breast was considered to be TP if it was detected in either one or both views. FT: feature-based, NN: DCNN using transfer learning from both mammography and DBT mass images. 

Computer-aided detection of breast microcalcifications in digital breast tomosynthesis (DBT) volume.

Flow diagram illustrating our CADe systems for detection of clustered microcalcifications in 2D digital mammograms (DM), 3D digital breast tomosynthesis (DBT) and 2D planar projection (PPJ) generated from DBT. The dotted line separates the CADe systems and the blocks crossing the dotted line show methods common to the CADe systems in the columns. For example, the ‘CNN classifier’ block is used in both DM and PPJ CADe systems. Each block is specifically trained for each type of images although the general techniques are similar. 

CNN: convolution neural network

LDA: linear discriminant analysis

MCR: multiscale calcification response 

CNR: contrast-to-noise ratio

EMCR: enhanced modulated calcification response

Flow diagram of our joint-CADe system for improving cluster detection in DBT:  (left) cluster detection processes adapted from the individual CADDBT and CADPPJ systems. The two processes are performed independently although some steps are shown with the same block in the schematic to illustrate the similarity. (right) joint FP reduction process after pooling the detected clusters from the two paths shown on the left. DBT and PPJ represent clusters from the CADDBT and CADPPJ systems, respectively, and LDAi (i=1,2,3) are linear discriminant analysis classifiers trained for each path. 

Computer-aided detection of microcalcification in 2D and 3D breast images.