Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Disease Classification and Impact of Pretrained Deep Convolution Neural Networks on Diverse Medical Imaging Datasets across Imaging Modalities

Jutika Borah, Kumaresh Sarmah, Hidam Kumarjit Singh

TL;DR

A comprehensive performance analysis showed that the use of pretrained models as fixed feature extractors yields poor performance irrespective of the datasets, and found that deeper and more complex architectures did not necessarily result in the best performance.

Abstract

Imaging techniques such as Chest X-rays, whole slide images, and optical coherence tomography serve as the initial screening and detection for a wide variety of medical pulmonary and ophthalmic conditions respectively. This paper investigates the intricacies of using pretrained deep convolutional neural networks with transfer learning across diverse medical imaging datasets with varying modalities for binary and multiclass classification. We conducted a comprehensive performance analysis with ten network architectures and model families each with pretraining and random initialization. Our finding showed that the use of pretrained models as fixed feature extractors yields poor performance irrespective of the datasets. Contrary, histopathology microscopy whole slide images have better performance. It is also found that deeper and more complex architectures did not necessarily result in the best performance. This observation implies that the improvements in ImageNet are not parallel to the medical imaging tasks. Within a medical domain, the performance of the network architectures varies within model families with shifts in datasets. This indicates that the performance of models within a specific modality may not be conclusive for another modality within the same domain. This study provides a deeper understanding of the applications of deep learning techniques in medical imaging and highlights the impact of pretrained networks across different medical imaging datasets under five different experimental settings.

Disease Classification and Impact of Pretrained Deep Convolution Neural Networks on Diverse Medical Imaging Datasets across Imaging Modalities

TL;DR

A comprehensive performance analysis showed that the use of pretrained models as fixed feature extractors yields poor performance irrespective of the datasets, and found that deeper and more complex architectures did not necessarily result in the best performance.

Abstract

Imaging techniques such as Chest X-rays, whole slide images, and optical coherence tomography serve as the initial screening and detection for a wide variety of medical pulmonary and ophthalmic conditions respectively. This paper investigates the intricacies of using pretrained deep convolutional neural networks with transfer learning across diverse medical imaging datasets with varying modalities for binary and multiclass classification. We conducted a comprehensive performance analysis with ten network architectures and model families each with pretraining and random initialization. Our finding showed that the use of pretrained models as fixed feature extractors yields poor performance irrespective of the datasets. Contrary, histopathology microscopy whole slide images have better performance. It is also found that deeper and more complex architectures did not necessarily result in the best performance. This observation implies that the improvements in ImageNet are not parallel to the medical imaging tasks. Within a medical domain, the performance of the network architectures varies within model families with shifts in datasets. This indicates that the performance of models within a specific modality may not be conclusive for another modality within the same domain. This study provides a deeper understanding of the applications of deep learning techniques in medical imaging and highlights the impact of pretrained networks across different medical imaging datasets under five different experimental settings.
Paper Structure (13 sections, 3 figures, 1 table)

This paper contains 13 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Materials and Methodology
  3. Datasets
  4. Data Preparation and Preprocessing
  5. Model Architectures
  6. Experimental Settings, Training Strategies and Performance Evaluation
  7. Experimental Results and Analysis
  8. Evaluation with Pretraining and Random Initialization
  9. Evaluation with Pretrained DCNN architectures
  10. Evaluation with Pretrained DCNN architectural family
  11. Discussion
  12. Challenges And Future Prospects
  13. Conclusion

Figures (3)

  • Figure 1: Image from top (a) Normal CXR, (b) Bacterial Pneumonia CXR, and (c) Viral Pneumonia CXR. Image at the middle (a) Choroidal neovascularization (CNV), (b) Diabetic macular edema (DME), (c) Drusen, and (d) Normal. Image at the top, WSI Histopathology (a) Lung adenocarcinoma, (b) Lung Normal, and (c) Lung squamous cell carcinoma.
  • Figure 2: Block representation of experimental design and evaluation framework. The figure shows all five different settings for performance evaluations with pretraining and random initializations.
  • Figure 3: Visual summary in scatter plots showing AUCs of (a) CXR (b) OCT, and (c) WSI of all DCNN pretrained models with pretraining and random initialization. The five different settings with the number of parameters represent the complexity of the models (VGG (16, 19), ResNet (50, 101, 152), Inceptionv3, Inception-ResNet-v2, Xception, and DenseNet (121, 201)). The values are represented in a logarithm scale (Table \ref{['tab1']}).