Deep learning–based histologic classifiers enable molecular subtyping of metastatic prostate cancer
Zhijun Chen, Erolcan Sayar, Daniela Guevara, Helen Richards, Haoyue Zhang, Radhika A. Patel, Agnes C. Gawne, Lucas J. Liu, Ilsa Coleman, Ruth Dumpit, Colm Morrissey, Michael T. Schweizer, Ruben Raychaudhuri, Laura S. Graham, Evan Y. Yu, Heather H. Cheng, Chien-Kuang C. Ding, Yuzhuo Wang, Peter Choyke, Baris Turkbey, Chantal Chanel-Vos, Christina Fedorov, John R. Otilano, Troy Kane, Jyothi Manohar, Michael Sigouros, Jones T. Nauseef, Ana Molina, David Nanus, Scott T. Tagawa, Juan Miguel Mosquera, Himisha P. Beltran, Ruth Etzioni, Peter S. Nelson, Rama Soundararajan, Ana M. Aparicio, Cora N. Sternberg, Michael C. Haffner, Stephanie A. Harmon
Zhijun Chen, Erolcan Sayar, Daniela Guevara, Helen Richards, Haoyue Zhang, Radhika A. Patel, Agnes C. Gawne, Lucas J. Liu, Ilsa Coleman, Ruth Dumpit, Colm Morrissey, Michael T. Schweizer, Ruben Raychaudhuri, Laura S. Graham, Evan Y. Yu, Heather H. Cheng, Chien-Kuang C. Ding, Yuzhuo Wang, Peter Choyke, Baris Turkbey, Chantal Chanel-Vos, Christina Fedorov, John R. Otilano, Troy Kane, Jyothi Manohar, Michael Sigouros, Jones T. Nauseef, Ana Molina, David Nanus, Scott T. Tagawa, Juan Miguel Mosquera, Himisha P. Beltran, Ruth Etzioni, Peter S. Nelson, Rama Soundararajan, Ana M. Aparicio, Cora N. Sternberg, Michael C. Haffner, Stephanie A. Harmon
View: Text | PDF
Research Article Clinical Research Oncology

Deep learning–based histologic classifiers enable molecular subtyping of metastatic prostate cancer

Abstract

Metastatic prostate cancer is a clinically and molecularly heterogeneous disease. Under the selective pressure of androgen receptor–directed (AR-directed) therapies, resistant phenotypes frequently emerge, posing significant diagnostic and therapeutic challenges. Neuroendocrine prostate cancer (NEPC) is a clinically important phenotype characterized by lineage plasticity, neuroendocrine features, visceral metastases, and poor prognosis. Accurate diagnosis of NEPC remains difficult owing to its histologic and molecular complexity but has high clinical relevance. In this study, we developed a deep learning model that leverages interpretable cellular features to improve feature extraction from H&E-stained tissue sections (NEURAL-PC). By incorporating a multiple-instance learning framework, NEURAL-PC enables robust NEPC classification solely from H&E tumor images, achieving an area under the receiver operating characteristic curve of 0.921 in independent external validation. In addition to its diagnostic utility, NEURAL-PC provides prognostic information that enables further subclassification of advanced prostate cancer across diverse datasets, supporting its strong prognostic value and generalizability. Broadly, our work highlights a hybrid approach that integrates features across different domains, offering a promising strategy for developing reliable deep learning tools in pathology. Built on this framework, NEURAL-PC represents an extensively validated diagnostic and prognostic model for advanced prostate cancer.

Authors

Zhijun Chen, Erolcan Sayar, Daniela Guevara, Helen Richards, Haoyue Zhang, Radhika A. Patel, Agnes C. Gawne, Lucas J. Liu, Ilsa Coleman, Ruth Dumpit, Colm Morrissey, Michael T. Schweizer, Ruben Raychaudhuri, Laura S. Graham, Evan Y. Yu, Heather H. Cheng, Chien-Kuang C. Ding, Yuzhuo Wang, Peter Choyke, Baris Turkbey, Chantal Chanel-Vos, Christina Fedorov, John R. Otilano, Troy Kane, Jyothi Manohar, Michael Sigouros, Jones T. Nauseef, Ana Molina, David Nanus, Scott T. Tagawa, Juan Miguel Mosquera, Himisha P. Beltran, Ruth Etzioni, Peter S. Nelson, Rama Soundararajan, Ana M. Aparicio, Cora N. Sternberg, Michael C. Haffner, Stephanie A. Harmon

×

Figure 1

The workflow of the proposed NEURAL-PC model.

Options: View larger image (or click on image) Download as PowerPoint
The workflow of the proposed NEURAL-PC model. (A) Following initial trai...
(A) Following initial training from a subset of manually annotated images, a preliminary cancer detection algorithm was applied to each WSI, resulting in a probability map output and binary segmentation mask, which was then converted to GeoJSON format for import to QuPath and manual refinement by pathologists. General modifications to reduce false positives were then used for retraining the cancer detection model. The final model results in highly specific segmentation of cancerous regions, which performs well on various tissue sources. (B) Tiles of size 250 × 250 at ×40 magnification with more than 10% cancer area were selected, nuclei were segmented using the Cellpose (48) algorithm, and the cellular feature (59) vectors corresponding to these tiles were extracted. (C) In the unsupervised pre-training step, the model was trained to reconstruct tiles as well as the cellular features. Tiles are transformed by an encoder, f, and a projector, p, into latent vectors zm and zc; then a decoder, gC, uses only zc to reconstruct the cellular feature vector corresponding to the input tile, and another decoder, gI, uses the sum z = zm + zc to reconstruct the input tile image. The objective is to minimize the differences between the input tile image and the reconstructed tile image, as well as the difference between the input cellular features and the reconstructed cellular features (see Methods). (D) The trained encoder f and projector p were combined with a multiple-instance learning classifier h to aggregate all tile features corresponding to an input image to produce a core/slide-level NEPC prediction. Digital WSIs were originally acquired at 0.25 μm/pixel (×40 magnification); scale bars are shown for each image.