Biologic and Prognostic Feature Scores from Whole-Slide Histology Images Using Deep Learning

Histopathology is a reflection of the molecular changes and provides prognostic phenotypes representing the disease progression. In this study, we introduced feature scores generated from hematoxylin and eosin histology images based on deep learning (DL) models developed for prostate pathology. We demonstrated that these feature scores were significantly prognostic for time to event endpoints (biochemical recurrence and cancer-specific survival) and had simultaneously molecular biologic associations to relevant genomic alterations and molecular subtypes using already trained DL models that were not previously exposed to the datasets of the current study. Further, we discussed the potential of such feature scores to improve the current tumor grading system and the challenges that are associated with tumor heterogeneity and the development of prognostic models from histology images. Our findings uncover the potential of feature scores from histology images as digital biomarkers in precision medicine and as an expanding utility for digital pathology.

[1]  P. Gumerlock,et al.  Human androgen receptor expression in prostate cancer following androgen ablation. , 1997, European urology.

[2]  S. Varambally,et al.  Induced Chromosomal Proximity and Gene Fusions in Prostate Cancer , 2009, Science.

[3]  P. Mahadevan,et al.  An overview , 2007, Journal of Biosciences.

[4]  A. Nakagawara,et al.  Role of p53 in Cell Death and Human Cancers , 2011, Cancers.

[5]  Ziding Feng,et al.  Prostate cancer specific mortality and Gleason 7 disease differences in prostate cancer outcomes between cases with Gleason 4 + 3 and Gleason 3 + 4 tumors in a population based cohort. , 2009, The Journal of urology.

[6]  Hilla Peretz,et al.  Ju n 20 03 Schrödinger ’ s Cat : The rules of engagement , 2003 .

[7]  J. Tchinda,et al.  Recurrent fusion of TMPRSS2 and ETS transcription factor genes in prostate cancer. , 2006, Science.

[8]  Jie Zhang,et al.  Nuclear Receptor-Induced Chromosomal Proximity and DNA Breaks Underlie Specific Translocations in Cancer , 2009, Cell.

[9]  A. Sivachenko,et al.  Exome sequencing identifies recurrent SPOP, FOXA1 and MED12 mutations in prostate cancer , 2012, Nature Genetics.

[10]  W. Isaacs,et al.  AR-V7 and resistance to enzalutamide and abiraterone in prostate cancer. , 2014, The New England journal of medicine.

[11]  J. Concato,et al.  A simulation study of the number of events per variable in logistic regression analysis. , 1996, Journal of clinical epidemiology.

[12]  Jürgen Schmidhuber,et al.  Deep learning in neural networks: An overview , 2014, Neural Networks.

[13]  Chris Sander,et al.  Copy number alteration burden predicts prostate cancer relapse , 2014, Proceedings of the National Academy of Sciences.

[14]  R. Rittmaster,et al.  Relative potency of testosterone and dihydrotestosterone in preventing atrophy and apoptosis in the prostate of the castrated rat. , 1996, The Journal of clinical investigation.

[15]  J. Köllermann,et al.  Combined histoarchitectural and cytological biopsy grading improves grading accuracy in low‐grade prostate cancer , 2012, International journal of urology : official journal of the Japanese Urological Association.

[16]  C S Song,et al.  Regulation of androgen action. , 1999, Vitamins and hormones.

[17]  M. Rubin,et al.  SPOP Mutation Drives Prostate Tumorigenesis In Vivo through Coordinate Regulation of PI3K/mTOR and AR Signaling. , 2017, Cancer cell.

[18]  D. Gleason Classification of prostatic carcinomas. , 1966, Cancer chemotherapy reports.

[19]  F. Harrell,et al.  Regression models for prognostic prediction: advantages, problems, and suggested solutions. , 1985, Cancer treatment reports.

[20]  S. Varambally,et al.  Therapeutic Targeting of SPINK1-Positive Prostate Cancer , 2011, Science Translational Medicine.

[21]  M Soledad Cepeda,et al.  Comparison of logistic regression versus propensity score when the number of events is low and there are multiple confounders. , 2003, American journal of epidemiology.

[22]  C. Bangma,et al.  Disease‐specific death and metastasis do not occur in patients with Gleason score ≤6 at radical prostatectomy , 2015, BJU international.

[23]  D. Bostwick,et al.  Staging of early prostate cancer: a proposed tumor volume-based prognostic index. , 1993, Urology.

[24]  H. Zeng,et al.  The association between SPINK1 and clinical outcomes in patients with prostate cancer: a systematic review and meta-analysis , 2017, OncoTargets and therapy.

[25]  Rajvir Dahiya,et al.  Hormonal, cellular, and molecular control of prostatic development. , 2003, Developmental biology.

[26]  N. Kyprianou,et al.  Activation of programmed cell death in the rat ventral prostate after castration. , 1988, Endocrinology.

[27]  G. Kristiansen,et al.  Konsenskonferenz 2014 der ISUP zur Gleason-Graduierung des Prostatakarzinoms , 2016, Der Pathologe.

[28]  Guigang Zhang,et al.  Deep Learning , 2016, Int. J. Semantic Comput..

[29]  Ewout W Steyerberg,et al.  Internal and external validation of predictive models: a simulation study of bias and precision in small samples. , 2003, Journal of clinical epidemiology.

[30]  H. Zeng,et al.  The association between SPINK1 and clinical outcomes in patients with prostate cancer: a systematic review and meta-analysis , 2017, OncoTargets and therapy.

[31]  Gleason Df Classification of prostatic carcinomas. , 1966 .

[32]  M. Delgado-Rodríguez,et al.  Systematic review and meta-analysis. , 2017, Medicina intensiva.

[33]  Yao-Tseng Chen,et al.  Gene fusions between TMPRSS2 and ETS family genes in prostate cancer: frequency and transcript variant analysis by RT-PCR and FISH on paraffin-embedded tissues , 2007, Modern Pathology.

[34]  N. Razavian,et al.  Classification and mutation prediction from non–small cell lung cancer histopathology images using deep learning , 2018, Nature Medicine.

[35]  D. Brat,et al.  Predicting cancer outcomes from histology and genomics using convolutional networks , 2017, Proceedings of the National Academy of Sciences.

[36]  T. Wheeler,et al.  Moving Beyond Gleason Scoring. , 2019, Archives of pathology & laboratory medicine.

[37]  B. Delahunt,et al.  International Society of Urological Pathology (ISUP) grading of prostate cancer – An ISUP consensus on contemporary grading , 2016, APMIS : acta pathologica, microbiologica, et immunologica Scandinavica.

[38]  T. Stamey,et al.  Morphologic and clinical significance of multifocal prostate cancers in radical prostatectomy specimens. , 2002, Urology.

[39]  Ce Zhang,et al.  Predicting non-small cell lung cancer prognosis by fully automated microscopic pathology image features , 2016, Nature Communications.

[40]  A. Uitterlinden,et al.  Novel taxonomy-independent deep learning microbiome approach allows for accurate classification of different forensically relevant human epithelial materials. , 2019, Forensic science international. Genetics.

[41]  S. Lowe,et al.  Tumor suppressive functions of p53. , 2009, Cold Spring Harbor perspectives in biology.

[42]  B. Delahunt,et al.  [The 2014 consensus conference of the ISUP on Gleason grading of prostatic carcinoma]. , 2016, Der Pathologe.

[43]  D. Jäger,et al.  Overexpression of nuclear AR-V7 protein in primary prostate cancer is an independent negative prognostic marker in men with high-risk disease receiving adjuvant therapy. , 2017, Urologic oncology.

[44]  Loris Nanni,et al.  Handcrafted vs. non-handcrafted features for computer vision classification , 2017, Pattern Recognit..

[45]  P. Pandolfi,et al.  SPOP Promotes Ubiquitination and Degradation of the ERG Oncoprotein to Suppress Prostate Cancer Progression. , 2015, Molecular cell.

[46]  Andrew J. Schaumberg,et al.  D R A F T H&E-stained Whole Slide Image Deep Learning Predicts SPOP Mutation State in Prostate Cancer , 2017 .

[47]  Guangtao Zhai,et al.  A Deep Learning-Based Radiomics Model for Prediction of Survival in Glioblastoma Multiforme , 2017, Scientific Reports.

[48]  J. López-Guerrero,et al.  Clinico-pathological significance of the molecular alterations of the SPOP gene in prostate cancer. , 2014, European journal of cancer.

[49]  Jun Luo,et al.  The mutational landscape of prostate cancer. , 2013, European urology.

[50]  B. Trock,et al.  SPINK1 Defines a Molecular Subtype of Prostate Cancer in Men with More Rapid Progression in an at Risk, Natural History Radical Prostatectomy Cohort. , 2016, The Journal of urology.

[51]  Steven J. M. Jones,et al.  The Molecular Taxonomy of Primary Prostate Cancer , 2015, Cell.

[52]  Jaya M Satagopan,et al.  TMPRSS2–ERG gene fusion is associated with low Gleason scores and not with high-grade morphological features , 2010, Modern Pathology.

[53]  Thomas Wiegel,et al.  Guidelines on Prostate Cancer , 2013 .

[54]  John T. Wei,et al.  The role of SPINK1 in ETS rearrangement-negative prostate cancers. , 2008, Cancer cell.

[55]  Mahul B Amin,et al.  Contemporary Gleason Grading of Prostatic Carcinoma: An Update With Discussion on Practical Issues to Implement the 2014 International Society of Urological Pathology (ISUP) Consensus Conference on Gleason Grading of Prostatic Carcinoma , 2017, The American journal of surgical pathology.

[56]  D. Tindall,et al.  Splicing of a novel androgen receptor exon generates a constitutively active androgen receptor that mediates prostate cancer therapy resistance. , 2008, Cancer research.

[57]  G. Coetzee,et al.  Contribution of the Androgen Receptor to Prostate Cancer Predisposition and Progression , 2004, Cancer and Metastasis Reviews.

[58]  L. Egevad,et al.  A Contemporary Prostate Cancer Grading System: A Validated Alternative to the Gleason Score. , 2016, European urology.

[59]  D. Tindall,et al.  Alternatively spliced androgen receptor variants. , 2011, Endocrine-related cancer.

[60]  B. Delahunt,et al.  The 2014 International Society of Urological Pathology (ISUP) Consensus Conference on Gleason Grading of Prostatic Carcinoma: Definition of Grading Patterns and Proposal for a New Grading System , 2015, The American journal of surgical pathology.

[61]  Ziding Feng,et al.  Histologic Grading of Prostatic Adenocarcinoma Can Be Further Optimized: Analysis of the Relative Prognostic Strength of Individual Architectural Patterns in 1275 Patients From the Canary Retrospective Cohort , 2016, The American journal of surgical pathology.

[62]  Andrew Zisserman,et al.  Very Deep Convolutional Networks for Large-Scale Image Recognition , 2014, ICLR.

[63]  C. Heinlein,et al.  Androgen receptor (AR) coregulators: an overview. , 2002, Endocrine reviews.