IMPROVED BREAST CANCER DIAGNOSIS AND PROGNOSIS BY COMUTATIONAL MODELING AND IMAGE ANALYSIS David E. Axelrod J.-A. Chapman, W.A. Christens-Barry, H.L. Lickley, N.A. Miller, J. Qian, L. Sontag, B. Subramanian, Y. Yuan NCI, NJCCR, Busch 111704
BREAST CANCER GOAL Clinical data Models Patient prognosis OUTLINE Breast cancer stages: in situ and invasive Clinical data Models Prediction Image analysis Prognosis
NORMAL BREAST ANATOMY Skarin, A.T. Breast Cancer I Slide Atlas of Diagnostic Oncology, Bristol-Myers Squibb Oncology
BREAST CANCER Skarin, A.T. Breast Cancer I Slide Atlas of Diagnostic Oncology, Bristol-Myers Squibb Oncology
BREAST TUMOR PROGRESSION Conventional View Normal Ductal Carcinoma In Situ Invasive Ductal Carcinoma Metastasis ( DCIS) (IDC) (M)
BREAST TUMOR PROGRESSION Conventional View Normal Ductal Carcinoma In Situ Invasive Ductal Carcinoma Metastasis ( DCIS) (IDC) (M) Normal Atypical Hyperplasia DCIS1 DCIS2 DCIS3 IDC1 IDC2 IDC3 M (AH)
DUCTAL CARCINOMA IN SITU IMPORTANCE OF GRADING DUCTAL CARCINOMA IN SITU 220,000 Breast Cancers / year 20% DCIS 32% recurrence free DCIS outcome 68% recur (DCIS or IDC) DCIS heterogeneity: 25% intermediate grade 50% mixed grades
PROGNOSIS BY PATHOLOGIST Miller, N.A. et al. The Breast Journal 7: 292-302 (2001) Nuclear grade No. Recurrence Recurrence Worst DCIS Invasive Grade 1 1 0 0 Grade 2 35 4 6 Grade 3 52 13 5 p = 0.18 p = 0.73 Conclude: Nuclear grade is not prognostic.
BREAST TUMOR PROGRESSION Conventional View Normal Ductal Carcinoma In Situ Invasive Ductal Carcinoma Metastasis ( DCIS) (IDC) (M) Normal Atypical Hyperplasia DCIS1 DCIS2 DCIS3 IDC1 IDC2 IDC3 M (AH)
CLINICAL OBSERVATIONS EXPECTATION DCIS1 DCIS2 DCIS3 IDC1 IDC2 IDC3 CLINICAL OBSERVATIONS Van Nuys Classification Holland Classification IDC DCIS IDC DCIS 1 2 3 1 2 3 1 90.10 26.73 11.88 1 65.66 53.54 12.12 2 55.45 87.13 55.45 2 27.27 117.17 57.58 3 3.96 25.74 141.58 3 4.04 23.23 137.38 Sum of observations of Gupta, Cadman and Leong, normalized to 498.
BREAST TUMOR PROGRESSION Conventional View Normal Ductal Carcinoma In Situ Invasive Ductal Carcinoma Metastasis ( DCIS) (IDC) (M) Normal Atypical Hyperplasia DCIS1 DCIS2 DCIS3 IDC1 IDC2 IDC3 M (AH)
Mommers et al. J. Pathol. 194: 327-333 (2001)
Buerger et al. J. Pathol. 187; 396-402 (1999)
" Unless you can express your knowledge with numbers, your knowledge is meager and unsatisfactory." William Thompson Lord Kelvin 1824-1907 Smithsonian Institution of Washington, 1857
B. Subramanian and D.E. Axelrod Progression of Heterogeneous Breast Tumors J. Theoret. Biol. 210: 107-119 (2001) Purpose: Pathways for tumor progression (compartment models) Transition rates between compartments Data: Co-occurrence frequencies of DCIS and IDC Method: Genetic algorithm (GA) search for transition rates Result: GA can’t reproduce data with models Conclusion: GA and/or models not adequate
PROBLEMS Genetic algorithm limitations, stuck in local minimum Pathway models not describe the biological situation Polluted data combined data from five labs different criteria to classify grades
PROBLEMS SOLUTIONS Genetic algorithm 1. Directed search limitations, stuck in local minimum seed Nelder-Mead simplex Pathway models 2. New pathway not describe the biological situation relax assumption DCIS -> IDC Polluted data 3. Combine similar data combined data from five labs combine data from three labs different criteria to classify grades same criteria to classify grades
CLINICAL OBSERVATIONS Van Nuys Classification Holland Classification IDC DCIS IDC DCIS 1 2 3 1 2 3 1 90.10 26.73 11.88 1 65.66 53.54 12.12 2 55.45 87.13 55.45 2 27.27 117.17 57.58 3 3.96 25.74 141.58 3 4.04 23.23 137.38 Sum of observations of Gupta, Cadman and Leong, normalized to 498.
PATHWAYS Linear Nonlinear Branched
DIFFERENTIAL EQUATIONS
BRANCHED PATHWAY
PATHWAY SIMULATIONS Linear IDC DCIS 1 2 3 1 94.62 49.80 0 1 2 3 1 94.62 49.80 0 2 0 114.54 74.70 3 0 0 164.34 Non-linear IDC DCIS 1 2 3 1 60.00 0 0 2 84.00 120.00 78.00 3 0 0 156.00 Branched IDC DCIS 1 2 3 1 103.48 0 0 2 64.68 103.48 71.14 3 0 0 155.22
PATHWAY SIMULATIONS Linear IDC DCIS 1 2 3 1 94.62 49.80 0 1 2 3 1 94.62 49.80 0 2 0 114.54 74.70 3 0 0 164.34 Non-linear IDC DCIS 1 2 3 1 60.00 0 0 2 84.00 120.00 78.00 3 0 0 156.00 Observed - Van Nuys Classification IDC DCIS 1 2 3 1 90.10 26.73 11.88 2 55.45 87.13 55.45 3 3.96 25.74 141.58 Branched IDC DCIS 1 2 3 1 103.48 0 0 2 64.68 103.48 71.14 3 0 0 155.22
PATHWAYS Linear Nonlinear Branched Parallel
PARALLEL PATHWAY Common Progenitor
PARALLEL PATHWAY p (0.642) 11 12 13 21 22 23 31 32 33
PARALLEL PATHWAY p (0.642) p (0.326) 11 12 13 21 22 23 31 32 33 11 12 13 21 22 23 31 32 33 11 12 13 21 22 23 31 32 33
PARALLEL PATHWAY p (0.032) p (0.642) p (0.326) 11 12 13 21 22 23 11 12 13 21 22 23 31 32 33 11 12 13 21 22 23 31 32 33 11 12 13 21 22 23 31 32 33
PATHWAY SIMULATIONS Linear IDC DCIS 1 2 3 1 94.62 49.80 0 1 2 3 1 94.62 49.80 0 2 0 114.54 74.70 3 0 0 164.34 Non-linear IDC DCIS 1 2 3 1 60.00 0 0 2 84.00 120.00 78.00 3 0 0 156.00 Branched IDC DCIS 1 2 3 1 103.48 0 0 2 0 103.48 71.14 3 0 0 155.22 Parallel IDC DCIS 1 2 3 1 106.57 40.59 7.97 2 40.59 106.57 40.59 3 7.97 40.59 106.57
COMPARISON OF RESULTS Clinical Observation Model Simulation Van Nuys Classification Parallel Model IDC DCIS IDC DCIS 1 2 3 1 2 3 1 90.10 26.73 11.88 1 106.57 40.59 7.97 2 55.45 87.13 55.45 2 40.59 106.57 40.59 3 3.96 25.74 141.58 3 7.97 40.59 106.57
BREAST TUMOR PROGRESSION Conventional View - Linear Progression Normal Ductal Carcinoma In Situ Invasive Ductal Carcinoma Metastasis New View - Parallel Progression Ductal Carcinoma In Situ Common Progenitor Normal Metastasis Invasive Ductal Carcinoma
Evaluation of pathways for progression of heterogeneous breast tumors PARALLEL PATHWAY Common Progenitor L. Sontag and D. E. Axelrod Evaluation of pathways for progression of heterogeneous breast tumors J. Theoret. Biol. 232: 179-189 (2005)
They include data on diagnosis and prognosis Slides 34-45 are excluded. They include data on diagnosis and prognosis of breast ductal carcinoma in situ by image analysis which has been submitted for publication.
CONCLUSION GOAL: Clinical data Models Patient prognosis OUTCOME: Clinical data Models Improved Patient prognosis