1
|
Anton J, Castelli L, Chan MF, Outters M, Tang WH, Cheung V, Shukla P, Walambe R, Kotecha K. How Well Do Self-Supervised Models Transfer to Medical Imaging? J Imaging 2022; 8:jimaging8120320. [PMID: 36547485 PMCID: PMC9782186 DOI: 10.3390/jimaging8120320] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 10/01/2022] [Revised: 11/19/2022] [Accepted: 11/21/2022] [Indexed: 12/05/2022] Open
Abstract
Self-supervised learning approaches have seen success transferring between similar medical imaging datasets, however there has been no large scale attempt to compare the transferability of self-supervised models against each other on medical images. In this study, we compare the generalisability of seven self-supervised models, two of which were trained in-domain, against supervised baselines across eight different medical datasets. We find that ImageNet pretrained self-supervised models are more generalisable than their supervised counterparts, scoring up to 10% better on medical classification tasks. The two in-domain pretrained models outperformed other models by over 20% on in-domain tasks, however they suffered significant loss of accuracy on all other tasks. Our investigation of the feature representations suggests that this trend may be due to the models learning to focus too heavily on specific areas.
Collapse
Affiliation(s)
- Jonah Anton
- Department of Computing, Imperial College London, London SW7 2AZ, UK
| | - Liam Castelli
- Department of Computing, Imperial College London, London SW7 2AZ, UK
- Correspondence:
| | - Mun Fai Chan
- Department of Computing, Imperial College London, London SW7 2AZ, UK
| | - Mathilde Outters
- Department of Computing, Imperial College London, London SW7 2AZ, UK
| | - Wan Hee Tang
- Department of Computing, Imperial College London, London SW7 2AZ, UK
| | - Venus Cheung
- Department of Computing, Imperial College London, London SW7 2AZ, UK
| | - Pancham Shukla
- Department of Computing, Imperial College London, London SW7 2AZ, UK
| | - Rahee Walambe
- Symbiosis Institute of Technology, Symbiosis International University, Pune 412115, India
| | - Ketan Kotecha
- Symbiosis Centre for Applied Artificial Intelligence, Symbiosis International University, Pune 412115, India
| |
Collapse
|
2
|
Feridouni Khamaneh Y, Kiani P, Miller RJD, Schlüter H, Friedrich RE. Complementing the pulp proteome via sampling with a picosecond infrared laser ( PIRL). Clin Oral Investig 2021; 25:6757-6768. [PMID: 33977388 PMCID: PMC8602158 DOI: 10.1007/s00784-021-03962-0] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 12/09/2020] [Accepted: 04/20/2021] [Indexed: 11/27/2022]
Abstract
OBJECTIVES The aim of this investigation was the detailed analysis of the human pulp proteome using the new picosecond infrared laser (PIRL)-based sampling technique, which is based on a completely different mechanism compared to mechanical sampling. Proteome analysis of healthy pulp can provide data to define changes in the proteome associated with dental disease. MATERIAL AND METHODS Immediately after extraction of the entire, undamaged tooth, 15 wisdom teeth were deep frozen in liquid nitrogen and preserved at -80°C. Teeth were crushed, and the excised frozen pulps were conditioned for further analysis. The pulps were sampled using PIRL, and the aspirates digested with trypsin and analyzed with mass spectrometry. Pulp proteins were categorized according to their gene ontology terminus. Proteins identified exclusively in this study were searched in the Human Protein Atlas (HPA) for gaining information about the main known localization and function. RESULTS A total of 1348 proteins were identified in this study. The comparison with prior studies showed a match of 72%. Twenty-eight percent of the proteins were identified exclusively in this study. Considering HPA, almost half of these proteins were assigned to tissues that could be pulp specific. CONCLUSION PIRL is releasing proteins from the dental pulp which are not dissolved by conventional sampling techniques. Clinical Relevance The presented data extend current knowledge on dental pulp proteomics in healthy teeth and can serve as a reference for studies on pulp proteomics in dental disease.
Collapse
Affiliation(s)
- Yaghoup Feridouni Khamaneh
- Department of Oral and Maxillofacial Surgery, University Medical Center Hamburg-Eppendorf, University of Hamburg, Hamburg, Germany.
- Institute of Clinical Chemistry and Laboratory Medicine, University Medical Center Hamburg-Eppendorf, University of Hamburg, Hamburg, Germany.
- Department of Periodontics, Preventive and Restorative Dentistry, University Medical Center Hamburg-Eppendorf, University of Hamburg, Hamburg, Germany.
- DMD, Dental Clinic Zahnvitalis, Julius-Vosseler-Str. 42, D-22527, Hamburg, Germany.
| | - Parnian Kiani
- Institute of Clinical Chemistry and Laboratory Medicine, University Medical Center Hamburg-Eppendorf, University of Hamburg, Hamburg, Germany
- Department of Chemistry, Lash Miller Chemical Laboratories, University of Toronto, 80 St. George Street, Toronto, ON, M5S 3H6, Canada
| | - R J Dwayne Miller
- Institute of Clinical Chemistry and Laboratory Medicine, University Medical Center Hamburg-Eppendorf, University of Hamburg, Hamburg, Germany
- Department of Chemistry, Lash Miller Chemical Laboratories, University of Toronto, 80 St. George Street, Toronto, ON, M5S 3H6, Canada
| | - Hartmut Schlüter
- Institute of Clinical Chemistry and Laboratory Medicine, University Medical Center Hamburg-Eppendorf, University of Hamburg, Hamburg, Germany
| | - Reinhard E Friedrich
- Department of Oral and Maxillofacial Surgery, University Medical Center Hamburg-Eppendorf, University of Hamburg, Hamburg, Germany
| |
Collapse
|
3
|
Wurlitzer M, Hessling E, Rinas K, Fuh M, Petersen H, Ricklefs F, Lamszus K, Regelsberger J, Maier S, Kruber S, Hansen NO, Miller RJD, Schlüter H. Mass Spectrometric Lipid Profiles of Picosecond Infrared Laser-Generated Tissue Aerosols Discriminate Different Brain Tissues. Lasers Surg Med 2019; 52:228-234. [PMID: 31067361 DOI: 10.1002/lsm.23096] [Citation(s) in RCA: 3] [Impact Index Per Article: 0.6] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Accepted: 04/14/2019] [Indexed: 11/05/2022]
Abstract
BACKGROUND AND OBJECTIVES A picosecond infrared laser (PIRL) has recently been demonstrated to cut biological tissue without scar formation based on the minimal destructive action on the surrounding cells. During cutting with PIRL, the irradiated tissue is ablated by a cold vaporization process termed desorption by impulsive vibrational excitation. In the resulting aerosol, all molecules are dissolved in small droplets and even labile biomolecules like proteins remain intact after ablation. It is hypothesized that these properties enable the PIRL in combination with mass spectrometry as an intelligent laser scalpel for guided surgery. In this study, it was tested if PIRL-generated tissue aerosols are applicable for direct analysis with mass spectrometry, and if the acquired mass spectra can be used to discriminate different brain areas. MATERIALS AND METHODS Brain tissues were irradiated with PIRL. The aerosols were collected and directly infused into a mass spectrometer via electrospray ionization without any sample preparation or lipid extraction. RESULTS The laser produced clear cuts with no marks of burning. Lipids from five different classes were identified in the mass spectra of all samples. By principal component analysis the different brain areas were clearly distinguishable from each other. CONCLUSIONS The results demonstrate the potential for real-time analysis of lipids with a PIRL-based laser scalpel, coupled to a mass spectrometer, for the discrimination of tissues during surgeries. Lasers Surg. Med. © 2019 Wiley Periodicals, Inc.
Collapse
Affiliation(s)
- Marcus Wurlitzer
- Department of Mass Spectrometric Proteomics, Institute for Clinical Chemistry and Laboratory Medicine, University Medical Center Hamburg-Eppendorf, Martinistr. 52, 20246 Hamburg, Germany
| | - Elisabeth Hessling
- Department of Mass Spectrometric Proteomics, Institute for Clinical Chemistry and Laboratory Medicine, University Medical Center Hamburg-Eppendorf, Martinistr. 52, 20246 Hamburg, Germany
| | - Karsten Rinas
- Department of Mass Spectrometric Proteomics, Institute for Clinical Chemistry and Laboratory Medicine, University Medical Center Hamburg-Eppendorf, Martinistr. 52, 20246 Hamburg, Germany
| | - MarcelineManka Fuh
- Department of Mass Spectrometric Proteomics, Institute for Clinical Chemistry and Laboratory Medicine, University Medical Center Hamburg-Eppendorf, Martinistr. 52, 20246 Hamburg, Germany
| | - Hannes Petersen
- Department of Otorhinolaryngology, Head and Neck Surgery and Oncology, University Medical Center Hamburg-Eppendorf, Martinistraße 52, 20246 Hamburg, Germany
| | - Franz Ricklefs
- Department of Neurosurgery, University Medical Center Hamburg-Eppendorf, Martinistraße 52, 20246 Hamburg, Germany
| | - Katrin Lamszus
- Department of Neurosurgery, University Medical Center Hamburg-Eppendorf, Martinistraße 52, 20246 Hamburg, Germany
| | - Jan Regelsberger
- Department of Neurosurgery, University Medical Center Hamburg-Eppendorf, Martinistraße 52, 20246 Hamburg, Germany
| | - Stephanie Maier
- Atomically Resolved Dynamics Division, Max Planck Institute for the Structure and Dynamics of Matter, Luruper Chaussee 149, 22761 Hamburg, Germany
| | - Sebastian Kruber
- Atomically Resolved Dynamics Division, Max Planck Institute for the Structure and Dynamics of Matter, Luruper Chaussee 149, 22761 Hamburg, Germany
| | - Nils-Owe Hansen
- Atomically Resolved Dynamics Division, Max Planck Institute for the Structure and Dynamics of Matter, Luruper Chaussee 149, 22761 Hamburg, Germany
| | - RJDwayne Miller
- Atomically Resolved Dynamics Division, Max Planck Institute for the Structure and Dynamics of Matter, Luruper Chaussee 149, 22761 Hamburg, Germany.,Departments of Chemistry and Physics, Lash Miller Chemical Laboratories, University of Toronto, 80 St. George Street, LM245A, Toronto, Ontario, M5S 3H6, Canada
| | - Hartmut Schlüter
- Department of Mass Spectrometric Proteomics, Institute for Clinical Chemistry and Laboratory Medicine, University Medical Center Hamburg-Eppendorf, Martinistr. 52, 20246 Hamburg, Germany
| |
Collapse
|
4
|
Hänel L, Kwiatkowski M, Heikaus L, Schlüter H. Mass spectrometry-based intraoperative tumor diagnostics. Future Sci OA 2019; 5:FSO373. [PMID: 30906569 DOI: 10.4155/fsoa-2018-0087] [Citation(s) in RCA: 36] [Impact Index Per Article: 7.2] [Reference Citation Analysis] [What about the content of this article? (0)] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/21/2018] [Accepted: 01/08/2019] [Indexed: 02/08/2023] Open
Abstract
In surgical oncology, decisions regarding the amount of tissue to be removed can have important consequences: the decision between preserving sufficient healthy tissue and eliminating all tumor cells is one to be made intraoperatively. This review discusses the latest technical innovations for a more accurate tumor margin localization based on mass spectrometry. Highlighting the latest mass spectrometric inventions, real-time diagnosis seems to be within reach; focusing on the intelligent knife, desorption electrospray ionization, picosecond infrared laser and MasSpec pen, the current technical status is evaluated critically concerning its scientific and medical practice.
Collapse
|