Machine Learning in Medical Images

We have proposed a deep learning system to provide automated PICC course and tip detection. The predicted location of PICC tip is 4.66 mm from ground truth on average with a standard deviation of 2.8 mm.

Organ Dose Estimation
ML powered automatic organ classifier to estimate organ specific radiation dose. Our preliminary results reveal higher than 96% accuracy in mapping of organs to phantom, providing high quality organ-specific radiation dose estimates at low cost.

Development of a clinically beneficial, controllable, and scalable knowledge extraction system to learn medical image features, disease ontology, body system, and image formation simultaneously.

Bone Age
A fully automated deep learning pipeline to segment a region of interest, standardize and preprocess input radiographs, and perform Bone Age Assessment. Our system  achieved average BAA accuracy of 98.56% within two years and 92.29% within one year.

Segmentation of skeletal muscle cross sectional area (CSA) on an axial computed tomography image taken at the third lumbar vertebra. Our best model, fine-tuned on 250 training images and ground-truth labels, achieves 0.930.02 Dice Similarity Coefficient (DSC) and 3.682.29% difference between predicted and ground truth muscle CSA on 150 held-out test cases.

Snowball Sampling
System can generate a large labeled dataset from a small initial training set using an iterative snowball sampling scheme. This sampling method selects candidates, classifies them by the deep convolutional neural network (DCNN), and then fully refines them by learning features from a Variational Autoencoder (VAE) and clustering the features by Gaussian Mixtures Models (GMM).

Classification of the Hemorrhage in three groups: Normal, Subarachnoid Hemorrhage, and Intraventricular Hemorrhage

Breast Cancer
Detection of breast cancer from mammograms.

Kidney Stones
Detecting kidney stone location and size.

Detecting the placement of ET tube.

Tuberculosis (TB)