Review — An Efficient Deep Neural Network Based Abnormality Detection and Multi‑Class Breast Tumor Classification

BUS-CNN, VGG-19, and YOLOv3, are Evaluated

Sik-Ho Tsang
3 min readNov 21, 2022
Breast Tumor Classification Datasets

An Efficient Deep Neural Network Based Abnormality Detection and Multi‑Class Breast Tumor Classification, BUS-CNN, by Dr. A.P.J.Abdul Kalam Technical University, 2022 JMTA (Sik-Ho Tsang @ Medium)
Medical Imaging, Medical Image Analysis, Image Classification

  • Breast ultrasound image datasets are trained on 3 deep-learning architectures: BUS-CNN, VGG-19, and YOLOv3.
  • YOLOv3 is found to be the best one.

Outline

  1. Three Deep-Learning Architectures
  2. Results

1. Three Deep-Learning Architectures

1.1. Framework

Framework
  • Standard deep learning workflow is presented.
  • The dataset is augmented, labeled and annotated. Then, it is split into training/validation/testing set. Then the training and validation sets are used for training the model. The testing set is used for evaluation.

1.2. Breast Ultrasound Convolution Neural Network (BUS-CNN)

Breast Ultrasound Convolution Neural Network (BUS-CNN)
  • BUS-CNN consists of six convolutional layers and three max pooling layers.
  • At the end, there are fully connected layers to sense the presence or absence of the breast tumor and its classification probability.

1.3. VGG-19

VGG-19
  • VGG-19 is evaluated. (Please feel free to read VGG-19 if interested.)

1.4. YOLOv3

YOLO-v3
  • YOLOv3 is evaluated. (Please feel free to read YOLOv3 if interested.)

2. Results

Receiver Operating Characteristics (ROC) of the trained deep learning models
  • YOLOv3 obtains the best ROC curve.
Analysis of different CNN models
  • YOLOv3 obtains the highest accuracy.

Though BUS-CNN is designed, VGG-19 and YOLOv3 are better.

--

--

Sik-Ho Tsang

PhD, Researcher. I share what I learn. :) Linktree: https://linktr.ee/shtsang for Twitter, LinkedIn, etc.