AI Outperforms Pathologists in Diagnosing Breast Cancer
By LabMedica International staff writers Posted on 20 Dec 2017 |
A study comparing the ability of Artificial Intelligence (AI) algorithms with expert pathologists in detecting metastatic breast cancer in whole-slide images found that the machine learning outperformed the pathologists. The results of the study published in the Journal of the American Medical Association suggests that deep learning algorithms have the ability to improve diagnosis and could be used to help clinicians detect cancer in the clinic.
The study pitted 11 pathologists with time constraints and one pathologist without time constraints against seven deep learning algorithms in analyzing a training data set of whole-slide images – 110 with and 160 without verified nodal metastases. Out of the 49 test slides with metastatic disease, the pathologists found 31 on an average, while the pathologist allowed to work without time constraint correctly identified 46 out of 49 slides with cancer and 79 out of 80 slides without cancer.
Among the seven deep learning algorithms, the best algorithm performed significantly better in the whole-slide image classification task as compared to the pathologists working with time constraints. The mean performance of the top five algorithms was comparable with that of the single pathologist working without time constraints. However, at a mean of 0.0125 false-positives per normal whole-slide image, the performance of the best-performing algorithm was comparable with that of the single pathologist working without time constraint.
The research was led by Babak Ehteshami Bejnordi, Radboud University Medical Centre Nijmegen in the Netherlands. The researchers concluded that while the findings suggested the potential utility of deep learning algorithms for pathological diagnosis, it required further assessment in a clinical setting.
The study pitted 11 pathologists with time constraints and one pathologist without time constraints against seven deep learning algorithms in analyzing a training data set of whole-slide images – 110 with and 160 without verified nodal metastases. Out of the 49 test slides with metastatic disease, the pathologists found 31 on an average, while the pathologist allowed to work without time constraint correctly identified 46 out of 49 slides with cancer and 79 out of 80 slides without cancer.
Among the seven deep learning algorithms, the best algorithm performed significantly better in the whole-slide image classification task as compared to the pathologists working with time constraints. The mean performance of the top five algorithms was comparable with that of the single pathologist working without time constraints. However, at a mean of 0.0125 false-positives per normal whole-slide image, the performance of the best-performing algorithm was comparable with that of the single pathologist working without time constraint.
The research was led by Babak Ehteshami Bejnordi, Radboud University Medical Centre Nijmegen in the Netherlands. The researchers concluded that while the findings suggested the potential utility of deep learning algorithms for pathological diagnosis, it required further assessment in a clinical setting.
Latest Industry News
- ECCMID Congress Name Changes to ESCMID Global
- Bosch and Randox Partner to Make Strategic Investment in Vivalytic Analysis Platform
- Siemens to Close Fast Track Diagnostics Business
- Beckman Coulter and Fujirebio Expand Partnership on Neurodegenerative Disease Diagnostics
- Sysmex and Hitachi Collaborate on Development of New Genetic Testing Systems
- Sysmex and CellaVision Expand Collaboration to Advance Hematology Solutions
- BD and Techcyte Collaborate on AI-Based Digital Cervical Cytology System for Pap Testing
- Medlab Middle East 2024 to Address Transformative Potential of Artificial Intelligence
- Seegene and Microsoft Collaborate to Realize a World Free from All Diseases and Future Pandemics
- Medlab Middle East 2024 to Highlight Importance of Sustainability in Laboratories
- Fujirebio and Agappe Collaborate on CLIA-Based Immunoassay
- Medlab Middle East 2024 to Highlight Groundbreaking NextGen Medicine
- bioMérieux Acquires Software Company LUMED to Support Fight against Antimicrobial Resistance
- Roche Acquires LumiraDx's Point of Care Technology for USD 295 Million
- Bruker Acquires IVD Manufacturer ELITech Group and Cell Imaging Company Phasefocus
- AMP Publishes Best Practice Guidance for Slice Testing Approach in Diagnostics