1 of 6

���Semantic Nuclei Segmentation for Medical Image Analysis using U-Net: An Exploration and Evaluation�

Presentation by:

Daniel Tweneboah Anyimadu

2018 DATA SCIENCE BOWL – NUCLEI SEGMENTATION CHALLENGE:

2 of 6

ABSTRACT

  • This study presents a comprehensive exploration and evaluation of a Semantic Nuclei Segmentation framework utilizing U-Net architecture for medical image analysis.
  • The primary goal is to automate nucleus detection in medical images, with potential implications for accelerating disease research and drug development.

2

  • The studies is carried out on the 2018 data science bowl dataset. The project involves data acquisition and exploration, preprocessing, U-Net architecture implementation for image segmentation, and performance evaluation.
  • Validation dataset results are assessed using intersection over union (IoU), average precision (AP) and accuracy.
  • The obtained results are as follows: IoU : 0.8624, mAP : 0.7746 and accuracy : 0.9668

3 of 6

Exploratory Data Analysis (EDA)

  • A visualization of Input & Mask (Groundtruth) Images :

3

4 of 6

U-Net Architecture Implementation

  • U-Net Architecture:
  • This is designed to automate the classification and segmentation of individual nucleus within input images containing multiple nuclei.

​

4

5 of 6

Comparative Analysis of Input Image, Groundtruth Mask & Predicted Mask

5

  • Training Images :
  • Validation Images :

6 of 6

PERFORMANCE EVALUATION OF U-NET MODEL

6

MODEL

IoU (TEST)

PRECISION (TEST)

mAP (TEST)

ACCURACY (TEST)

U-Net

0.8624

0.7746

0.7746

0.9668

  • Semantic nuclei segmentation with U-Net holds promise for transforming medical research as attested by the obtained results.