Download PDFOpen PDF in browser

Transfer Learning Pre-training Dataset Effect Analysis for Breast Cancer Imaging

8 pagesPublished: March 22, 2022

Abstract

Comparing with natural imaging datasets used in transfer learning, the effects of med- ical pre-training datasets are underexplored. In this study, we carry out transfer learning pre-training dataset effect analysis in breast cancer imaging by evaluating three popular deep neural networks and one patch-based convolutional neural network on three target datasets under different fine-tuning configurations. Through a series of comparisons, we conclude that the pre-training dataset, DDSM, is effective on two other mammogram datasets. However, it is ineffective on an ultrasound dataset. What is more, fine-tuning may mask the inefficacy of a pre-training dataset. In addition, the efficacy/inefficacy of DDSM on the target datasets is corroborated by a representational analysis. At last, we show that hybrid transfer learning cannot mitigate the masking effect of fine-tuning.

Keyphrases: breast cancer imaging, ddsm, fine tuning, pre training dataset, transfer learning

In: Hisham Al-Mubaid, Tamer Aldwairi and Oliver Eulenstein (editors). Proceedings of 14th International Conference on Bioinformatics and Computational Biology, vol 83, pages 108-115.

BibTeX entry
@inproceedings{BICOB2022:Transfer_Learning_Pre_training,
  author    = {Chanaka Bulathsinghalage and Lu Liu},
  title     = {Transfer Learning Pre-training Dataset Effect Analysis for Breast Cancer Imaging},
  booktitle = {Proceedings of 14th International Conference on Bioinformatics and Computational Biology},
  editor    = {Hisham Al-Mubaid and Tamer Aldwairi and Oliver Eulenstein},
  series    = {EPiC Series in Computing},
  volume    = {83},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2398-7340},
  url       = {/publications/paper/PXq1},
  doi       = {10.29007/nns7},
  pages     = {108-115},
  year      = {2022}}
Download PDFOpen PDF in browser