Format

Send to

Choose Destination
Clin Cancer Res. 2019 Oct 21. pii: clincanres.1495.2019. doi: 10.1158/1078-0432.CCR-19-1495. [Epub ahead of print]

Deep learning based on standard H&E images of primary melanoma tumors identifies patients at risk for visceral recurrence and death.

Author information

1
Department of Psychiatry, School of Medicine, NYU Langone Medical Center.
2
Department of Anesthesiology, Division of Perioperative Care and Pain Medicine, NYU Langone Medical Center.
3
Department of Medicine, Columbia University Irving Medical Center.
4
Department of Pediatrics, Columbia University Irving Medical Center.
5
Departments of Dermatology and Laboratory Medicine, University Hospitals Case Medical Center.
6
Vagelos College of Physicians and Surgeons, Columbia University Irving Medical Center.
7
Dermatology, Columbia University.
8
Smilow Cancer Center, Yale School of Medicine.
9
Department of Pathology, Yale School of Medicine.
10
Department of Oncology-Pathology, Karolinska Institute.
11
Department of Medicine, Division of Hematology/Oncology, Columbia University Irving Medical Center.
12
Department of Medicine, Jiaotong University School of Medicine.
13
Perlmutter Cancer Center, NYU Langone Health.
14
Dermatology and Pathology, Icahn School of Medicine at Mount Sinai.
15
Department of Pathology, University of British Columbia, Vancouver General Hospital.
16
Department of Psychiatry, Department of Neuroscience and Physiology, New York University School of Medicine.
17
Dermatology and Pathology, Geisinger Medical Center.
18
Department of Medicine, Division of Hematology/Oncology, Columbia University Irving Medical Center yms4@cumc.columbia.edu.

Abstract

PURPOSE:

Biomarkers for disease specific survival (DSS) in early stage melanoma are needed to select patients for adjuvant immunotherapy and accelerate clinical trial design. We present a pathology-based computational method using a deep neural network architecture for DSS prediction.

EXPERIMENTAL DESIGN:

The model was trained on 108 patients from four institutions and tested on 104 patients from Yale School of Medicine (YSM). A receiver operating characteristic (ROC) curve was generated based on vote aggregation of individual image sequences, an optimized cutoff was selected, and the computational model was tested on a third independent population of 51 patients from Geisinger Health Systems (GHS).

RESULTS:

Area under the curve (AUC) in the YSM patients was 0.905 (p<0.0001). AUC in the GHS patients was 0.880 (p<0.0001). Using the cutoff selected in the YSM cohort, the computational model predicted DSS in the GHS cohort based on Kaplan-Meier (KM) analysis (p<0.0001).

CONCLUSIONS:

The novel method presented is applicable to digital images, obviating the need for sample shipment and manipulation and representing a practical advance over current genetic and IHC-based methods.

Supplemental Content

Full text links

Icon for HighWire Icon for NYU School of Medicine
Loading ...
Support Center