Abstracts

RETURN TO ABSTRACT LISTING


1529. Deep Learning Prediction of Axillary Lymph Node Status From Primary Breast Cancer Ultrasound
Authors * Denotes Presenting Author
  1. Shawn Sun *; Columbia University Irving Medical Center/New York Presbyterian
  2. Simuyaki Mutasa; Thomas Jefferson University/Sidney Kimmel Medical College
  3. Michael Liu; Columbia University Irving Medical Center/New York Presbyterian
  4. Maham Siddique; Columbia University Irving Medical Center/New York Presbyterian
  5. Elise Desperito; Columbia University Irving Medical Center/New York Presbyterian
  6. Sachin Jambawalikar; Columbia University Irving Medical Center/New York Presbyterian
  7. Richard Ha; Columbia University Irving Medical Center/New York Presbyterian
Objective:
Detecting lymph node metastasis is an important part of staging and in the management of breast cancer patients. In this study, we investigate the ability of our convolutional neural network (CNN) to predict axillary lymph node metastasis using primary breast cancer ultrasound (US) images.

Materials and Methods:
An IRB approved study was performed. We collected 338 images (two orthogonal images) from 169 patients who underwent US evaluation at the time of breast cancer diagnosis from 2015-2018. These patients initially had suspiciously enlarged lymph nodes on US imaging and status was confirmed by tissue pathology. Of 169 patients, 64 patients had a positive lymph node for metastasis and 105 patients did not. A custom convolutional neural network trained from random initialization was utilized on 248 US images from 124 patients in the training and cross validation set and tested on 90 US images from 45 patients. The CNN was implemented entirely of 3 x 3 convolutional kernels and linear layers. The 9 convolutional kernels consisted entirely of 6 residual layers, totaling 12 convolutional layers. Feature maps were down-sampled using strided convolutions. Dropout with a 0.5 keep probability and L2 normalization was utilized. Training was implemented by using the Adam optimizer which performs parameter wise momentum augmented training. Network weights were initialized randomly. A final SoftMax score threshold of 0.5 from the average of raw logits from each pixel was used for two class classification (metastasis or no metastasis). Software code for this study was written in Python 3.6 using the TensorFlow module (1.130.10). Study was done on a Linux workstation with NVIDIA GTX 1070 Pascal GPU with 8 GB on chip memory, i7 CPU and 32 GB RAM. Model performance was assessed in terms of accuracy, sensitivity, specificity and area under the receiver operating curve (AUC).

Results:
Our CNN algorithm achieved an AUC of 0.72 (SD± 0.08) in predicting axillary lymph node metastasis from breast cancer US images in the testing dataset. The model had an accuracy of 72.6% (SD±8.4) with a sensitivity and specificity of 65.5% (SD± 28.6) and 78.9% (SD± 15.1) respectively.

Conclusion:
It’s feasible to predict axillary lymph node metastasis from breast cancer US images using a deep learning technique. This tool has the potential to aid in nodal staging in patients with breast cancer. Larger dataset will likely improve our prediction model and this can potentially be a non-invasive alternative to core needle biopsy and even sentinel lymph node evaluation.