External validation and transfer learning of convolutional neural networks for computed tomography dental artifact classification.

Phys Med Biol

Department of Medical Biophysics, University of Toronto, Toronto, Ontario, Canada. Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada. The Techna Institute for the Advancement of Technology for Health, Toronto, Ontario, Canada. Author to whom any correspondence should be addressed.

Published: February 2020

AI Article Synopsis

  • Implementing quality assurance for data in automated pipelines and image analysis helps prevent biases and misinterpretations.
  • This study validates the effectiveness of convolutional neural networks (CNNs) in detecting dental artifacts in head and neck CT images across multiple external datasets, showing that transfer learning enhances performance.
  • Results showed the highest accuracy (AUC of 0.92) when using larger resampling grids with transfer learning, while smaller grids or datasets did not yield better results.

Article Abstract

Quality assurance of data prior to use in automated pipelines and image analysis would assist in safeguarding against biases and incorrect interpretation of results. Automation of quality assurance steps would further improve robustness and efficiency of these methods, motivating widespread adoption of techniques. Previous work by our group demonstrated the ability of convolutional neural networks (CNN) to efficiently classify head and neck (H&N) computed-tomography (CT) images for the presence of dental artifacts (DA) that obscure visualization of structures and the accuracy of Hounsfield units. In this work we demonstrate the generalizability of our previous methodology by validating CNNs on six external datasets, and the potential benefits of transfer learning with fine-tuning on CNN performance. 2112 H&N CT images from seven institutions were scored as DA positive or negative. 1538 images from a single institution were used to train three CNNs with resampling grid sizes of 64, 128 and 256. The remaining six external datasets were used in five-fold cross-validation with a data split of 20% training/fine-tuning and 80% validation. The three pre-trained models were each validated using the five-folds of the six external datasets. The pre-trained models also underwent transfer learning with fine-tuning using the 20% training/fine-tuning data, and validated using the corresponding validation datasets. The highest micro-averaged AUC for our pre-trained models across all external datasets occurred with a resampling grid of 256 (AUC  =  0.91  ±  0.01). Transfer learning with fine-tuning improved generalizability when utilizing a resampling grid of 256 to a micro-averaged AUC of 0.92  ±  0.01. Despite these promising results, transfer learning did not improve AUC when utilizing small resampling grids or small datasets. Our work demonstrates the potential of our previously developed automated quality assurance methods to generalize to external datasets. Additionally, we showed that transfer learning with fine-tuning using small portions of external datasets can be used to fine-tune models for improved performance when large variations in images are present.

Download full-text PDF

Source
http://dx.doi.org/10.1088/1361-6560/ab63baDOI Listing

Publication Analysis

Top Keywords

transfer learning
24
external datasets
24
learning fine-tuning
16
quality assurance
12
resampling grid
12
pre-trained models
12
convolutional neural
8
neural networks
8
datasets
8
20% training/fine-tuning
8

Similar Publications

Introduction: Wearables are electronic devices worn on the body to collect health data. These devices, like smartwatches and patches, use sensors to gather information on various health parameters. This review highlights current use and the potential benefit of wearable technology in patients with inflammatory bowel disease (IBD).

View Article and Find Full Text PDF

Chronic obstructive pulmonary disease (COPD) is a prevalent chronic inflammatory airway disease with high incidence and significant disease burden. R-loops, functional chromatin structure formed during transcription, are closely associated with inflammation due to its aberrant formation. However, the role of R-loop regulators (RLRs) in COPD remains unclear.

View Article and Find Full Text PDF

This study investigates the electronic properties and photovoltaic (PV) performance of newly designed bithiophene-based dyes, focusing on their light harvesting efficiency (LHE), open-circuit voltage (V), fill factor (FF), and short-circuit current density (J).These new dyes are designed with the help of machine learning (ML) to design best donor acceptor designs. For this, we collect 2567 differenr electron donor groups and calculated their bandgap with the help of Random Forest (RF) Regression method.

View Article and Find Full Text PDF

IL-33, a neutrophil extracellular trap-related gene involved in the progression of diabetic kidney disease.

Inflamm Res

January 2025

Department of Nephrology, First Affiliated Hospital of Naval Medical University, Shanghai Changhai Hospital, Shanghai, China.

Background: Chronic inflammation is well recognized as a key factor related to renal function deterioration in patients with diabetic kidney disease (DKD). Neutrophil extracellular traps (NETs) play an important role in amplifying inflammation. With respect to NET-related genes, the aim of this study was to explore the mechanism of DKD progression and therefore identify potential intervention targets.

View Article and Find Full Text PDF

Soil microbiota plays crucial roles in maintaining the health, productivity, and nutrient cycling of terrestrial ecosystems. The persistence and prevalence of heterocyclic compounds in soil pose significant risks to soil health. However, understanding the links between heterocyclic compounds and microbial responses remains challenging due to the complexity of microbial communities and their various chemical structures.

View Article and Find Full Text PDF

Want AI Summaries of new PubMed Abstracts delivered to your In-box?

Enter search terms and have AI summaries delivered each week - change queries or unsubscribe any time!