System and method for accurate and rapid identification of diseased regions on biological images with applications to disease diagnosis and prognosis

A method for detecting biologically relevant structures in a hierarchical fashion comprising:

• a) identifying and segmenting gland lumens to produce segmented boundaries;
• b) extracting morphological features from the segmented boundaries;
• c) providing a Bayesian classifier to model the glandular area;
• d) assigning a value of probability of malignancy to each gland;
• e) classifying and labeling the gland as malignant if the probability exceeds the predetermined threshold τarea or labeling the gland as benign if the probability does not exceed the predetermined threshold τarea;
• f) using a MRF algorithm to refine the labeling of step(c).

A computer-based system for identifying and assessing the biological significance of abnormal regions in medical images comprising:

• a) segmenting and consolidating biologically relevant structures using the individual features of these structures, the features and classes of those structures that surround them, and the architectural arrangement of the structures;
• and
• b) analyzing the biologically relevant structures in a hierarchical manner using a Bayesian classifier to assess the biological significance of any abnormal regions;
• c) assigning the biological significance of step (b) to one of two classes by performing MRF iteration using weighted maximum posterior marginals estimation:
• those regions in which the available regions in which the available information is sufficient to make an assessment of the biological significance or those regions that should be examined at the next hierarchical level.

A computer-aided diagnosis system to automatically detect and grade the extent of LI in digitized HER2+ BC histopathology images comprising:

• a) identifying candidate image locations that may represent centers of lymphocytic nuclei using a region-growing algorithm to define sharp, well-defined boundaries;
• b) refining the initial lymphocyte identification of step (a) using a Maximum a Posteriori estimation to incorporate domain knowledge regarding lymphocyte size, luminance, and spatial proximity;
• c) further refining the detection result of step (b) using a Markov Random Field model to ensure that an object is more likely to be labeled as a lymphocyte nucleus if it surrounded by other lymphocyte nuclei;
• d) reiterating the Markov Random Field model of step (c) until convergence by using an Iterated Conditional Modes algorithm.

A method for detecting biologically relevant structures in a hierarchical fashion comprising:

• a) identifying and segmenting biologically relevant structures;
• b) calculating morphological or textural features for each biologically relevant structure;
• c) providing a Bayesian classifier to model the biologically relevant structure;
• d) classifying each biologically relevant structure using a MRF iteration using weighted maximum posterior marginals estimation;
• e) consolidating each biologically relevant structure into regions;
• f) drawing biological conclusions from consolidated regions.