FGAI4H-R-040-A12 Cambridge, 21-24 March 2023 Source: Nielsen
30 Slides8.20 MB
FGAI4H-R-040-A12 Cambridge, 21-24 March 2023 Source: Nielsen Santos Pereira Title: Att.12 – Presentation - How to determine the sample-size for machine learning in dental imaging Contact: Nielsen Santos Pereira Abstract: This PPT contains a presentation on how to determine the sample-size for machine learning in dental imaging given in the AI for Dentistry Symposium on 21 March 2023. E-mail: [email protected]
Meeting R - TG-Dental March 21-24 - 2023 Falk Schwendicke, Joachim Krois, Tarry Singh, Jae-Hong Lee, Akhilanand Chaurasia, Robert André Gaudin, Sergio Uribe, Hossein Mohammad-Rahimi, Janet Brinz, Anahita Haiat, Gürkan Ünsal, Nielsen Santos Pereira, Ulrike Kuchler, Shankeeth Vinayahalingam, Balazs Feher, Francesc Perez Pastor, Lisa Schneider, Chen Nadler, Sahel Hassanzadeh-Samani, Parisa Motie, Ragda Abdalla-Aslan, Teodora Karteva, Jelena Roganovic, Kunaal Dhingra, Prabhat Kumar Chaudhari, Olga Tryfonos, Marja Laine, Rata Rokhshad, Fatemeh Sohrabniya, Zeynab Pirayesh, Shada Alsalamah, Sakher AlQahtani, Revan Birke Koca-Ünsal, Lubaina T. Arsiwala, Parul Khare, Amit Punj, Manal Hamdan, Zaid Badr, Tamara Peric, Mihiri Silva, Bree Jones, Miroslav Radenković, Martha Duchrau, Mohammed Omar, Gowri Sivaramakrishnan, Jaisri Thoppay, Saujanya Karki, Tarja Tanner, Marja-Liisa Laitala, Johannes Tanne Dr. Nielsen S. Pereira – [email protected]
Question How to determine the samplesize for machine learning in dental imaging? Dr. Nielsen S. Pereira – [email protected]
Answers Dr. Nielsen S. Pereira – [email protected]
Answers Dr. Nielsen S. Pereira – [email protected]
Answers Dr. Nielsen S. Pereira – [email protected]
In Theory: Definition “In classical statistics, Sample-Size Determination Methodologies (SSDMs) estimate the optimum number of participants to arrive at scientifically valid results, often balancing an acceptable degree of precision with availability of resources” Sample-Size Determination Methodologies for Machine Learning in Medical Imaging Research: A Systematic Review September 2019 Canadian Association of Radiologists Journal 70(4) DOI:10.1016/j.carj.2019.06.002 Dr. Nielsen S. Pereira – [email protected]
In Theory: in Medicine/Dentistry “Analogously, for ML in medical imaging, we define an SSDM as a procedure to estimate the number of images required for an ML algorithm to reach a particular threshold of performance, or a sufficiently low generalizability error. While sample size issues may affect many ML disciplines, this is a particularly poignant challenge in medical imaging, where access to large quantities of high quality data is elusive” Sample-Size Determination Methodologies for Machine Learning in Medical Imaging Research: A Systematic Review September 2019 Canadian Association of Radiologists Journal 70(4) DOI:10.1016/j.carj.2019.06.002 Dr. Nielsen S. Pereira – [email protected]
In Theory: Approaches Sample-Size Determination Methodologies for Machine Learning in Medical Imaging Research: A Systematic Review September 2019 Canadian Association of Radiologists Journal 70(4) DOI:10.1016/j.carj.2019.06.002 Dr. Nielsen S. Pereira – [email protected]
In Practice The Goal of your Project The Task you are planning to The Architecture or Network The Availability of Data Previous researches or papers The Annotation Process The Infrastructure The Modality Dr. Nielsen S. Pereira – [email protected]
The Modalities Dr. Nielsen S. Pereira – [email protected]
The Modalities Dr. Nielsen S. Pereira – [email protected]
The Modalities Dr. Nielsen S. Pereira – [email protected]
The Modalities Dr. Nielsen S. Pereira – [email protected]
The Modalities Dr. Nielsen S. Pereira – [email protected]
The Modalities Dr. Nielsen S. Pereira – [email protected]
The Modalities Comparison Panoramic or OPG – 1.000 Dr. Nielsen S. Pereira – [email protected]
The Modalities Comparison Panoramic or OPG – 1.000 Periapical or Intraoral – 18.000 Dr. Nielsen S. Pereira – [email protected]
The Modalities Comparison Panoramic or OPG – 1.000 Periapical or Intraoral – 18.000 Lateral Cephalometric – 1.000 Dr. Nielsen S. Pereira – [email protected]
The Modalities Comparison Panoramic or OPG – 1.000 Periapical or Intraoral – 18.000 Lateral Cephalometric – 1.000 CBCT – 180.000 Dr. Nielsen S. Pereira – [email protected]
The DICOM Hierarchy Dr. Nielsen S. Pereira – [email protected]
Quantity x Quality MODEL Dr. Nielsen S. Pereira – [email protected]
Annotation Task 1. The Annotation Tool 2. Experienced and Certified Team 3. OMFR Supervisor 4. Annotation Guideline 5. Calibration Meeting 6. Calibration Dataset 7. Team annotation follow up and corrections 8. Specify a Deadline Dr. Nielsen S. Pereira – [email protected]
Annotation Task Dr. Nielsen S. Pereira – [email protected]
Example Dr. Nielsen S. Pereira – [email protected]
Open Dataset Risk 1. The Raw Data 2. Metadata (Gender, Sex, Equipment, etc.) 3. The ORIGIN of the data 4. Selection Criteria (Randomization / Anonymization 5. If annotated, the quality of the annotation Dr. Nielsen S. Pereira – [email protected]
Open Dataset Risk 1. The Raw Data 2. Metadata (Gender, Sex, Equipment, etc.) 3. The ORIGIN of the data 4. Selection Criteria (Randomization / Anonymization 5. If annotated, the quality of the annotation Dr. Nielsen S. Pereira – [email protected]
Open Dataset Risk 1. The Raw Data 2. Metadata (Gender, Sex, Equipment, etc.) 3. The ORIGIN of the data 4. Selection Criteria (Randomization / Anonymization 5. If annotated, the quality of the annotation Dr. Nielsen S. Pereira – [email protected]
Open Dataset Risk 1. The Raw Data 2. Metadata (Gender, Sex, Equipment, etc.) 3. The ORIGIN of the data 4. Selection Criteria (Randomization / Anonymization 5. If annotated, the quality of the annotation Dr. Nielsen S. Pereira – [email protected]
Thank you very much! Dr. Nielsen S. Pereira [email protected]
