Discussion on Medical Image ion & Evaluation Metrics

Course Q&A AI For Medicine AI for Medical Diagnosis
,
1

Before you post, please make sure to:

  1. Add a clear title that describes your question.

  2. Add all required tags in the tag option (top-right of the post):

    • The platform (e.g., dl-ai-learning-platform or Coursera-platform.)

    • The course week/module

  3. Write a detailed description of your issue.

Note: Screenshots are welcome, but please do not post full assignment code.

For more information, please check the FAQ category.

*Delete this message before posting

​Hi everyone,

​I’ve been reviewing the Soft Dice Loss formula L(P, G) = 1 - \frac{2 \sum p_i g_i}{\sum p_i^2 + \sum g_i^2} and comparing how different models perform on the same ground truth.

​It’s fascinating to see how background pixel predictions (where the ground truth is 0) heavily impact the denominator. Even if two models have identical true-positive prediction scores, the model with higher confidence in the background areas (closer to 0.5 instead of 0.3) ends up with a larger denominator, resulting in a higher loss.

​I would love to hear how others approach balancing background noise when optimizing segmentation models, especially when dealing with highly imbalanced medical datasets.

​Looking forward to your insights!

2

@Mohammed_Eltayeb1

my first question would is your query from ai for medicine specialisation? as you seem to have same query in multiple course categories.

if your query is not from ai for medicine specialisation, then move it to ai discussion.

3

Ok

4

@Mohammed_Eltayeb1

I am asking you this question because then we can address your query as I am confused between if you are having issue with assignment codes or is it a general query.