This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Computer applications, Bones, CT, CAD, CT-Quantitative, Computer Applications-Detection, diagnosis, Technical aspects, Computer Applications-General, Osteoporosis, Demineralisation-Bone
Authors:
K. S. Mader1, A. W. Sauter2, G. Sommer2, B. Stieltjes2; 1Zurich/CH, 2Basel/CH
DOI:
10.1594/ecr2017/C-1802
Methods and materials
For the study we used Thorax CT images from 15 different patients manually labeled by experts as being pathological or normal [1].
Morphological and texture information is extracted from the image and a model is trained to classify the vertebra based on the metrics.
For training the data were divided into two groups: training and validation.
The models were trained with the training data and only exposed to the validation data once the training was completed.
Since different models can deliver different results we used 4 different models using approaches of varying complexity and intrinsic interpretability: Random Forest,
Gradient Boosted Trees,
Decision Trees,
and Logistic Regression.
Local Interpretable Model-Agnostic Explanations (LIME) is applied to generate a list of relevant factors for a given classification [2].
A trained expert examines these outputs and assigns a feasibility score between 0 and 5 to each model.
The feasibility score is then compared to the validation score to determine if the interpretability is linked to the predictiveness.