High-performance prediction models for prostate cancer radiomics
When researchers are faced with building machine learning (ML) radiomic models, the first choice they have to make is what model to use. Naturally, the goal is to use the model with the best performance. But what is the best model? It is well known in ML that modern techniques such as gradient boost...
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2023-01-01
|
| Series: | Informatics in Medicine Unlocked |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2352914823000035 |
| _version_ | 1827999779699818496 |
|---|---|
| author | Lars Johannes Isaksson Marco Repetto Paul Eugene Summers Matteo Pepa Mattia Zaffaroni Maria Giulia Vincini Giulia Corrao Giovanni Carlo Mazzola Marco Rotondi Federica Bellerba Sara Raimondi Zaharudin Haron Sarah Alessi Paula Pricolo Francesco Alessandro Mistretta Stefano Luzzago Federica Cattani Gennaro Musi Ottavio De Cobelli Marta Cremonesi Roberto Orecchia Davide La Torre Giulia Marvaso Giuseppe Petralia Barbara Alicja Jereczek-Fossa |
| author_facet | Lars Johannes Isaksson Marco Repetto Paul Eugene Summers Matteo Pepa Mattia Zaffaroni Maria Giulia Vincini Giulia Corrao Giovanni Carlo Mazzola Marco Rotondi Federica Bellerba Sara Raimondi Zaharudin Haron Sarah Alessi Paula Pricolo Francesco Alessandro Mistretta Stefano Luzzago Federica Cattani Gennaro Musi Ottavio De Cobelli Marta Cremonesi Roberto Orecchia Davide La Torre Giulia Marvaso Giuseppe Petralia Barbara Alicja Jereczek-Fossa |
| author_sort | Lars Johannes Isaksson |
| collection | DOAJ |
| description | When researchers are faced with building machine learning (ML) radiomic models, the first choice they have to make is what model to use. Naturally, the goal is to use the model with the best performance. But what is the best model? It is well known in ML that modern techniques such as gradient boosting and deep learning have better capacity than traditional models to solve complex problems in high dimensions. Despite this, most radiomics researchers still do not focus on these models in their research. As access to high-quality and large data sets increase, these high-capacity ML models may become even more relevant. In this article, we use a large dataset of 949 prostate cancer patients to compare the performance of a few of the most promising ML models for tabular data: gradient-boosted decision trees (GBDTs), multilayer perceptions, convolutional neural networks, and transformers. To this end, we predict nine different prostate cancer pathology outcomes of clinical interest. Our goal is to give a rough overview of how these models compare against one another in a typical radiomics setting. We also investigate if multitask learning improves the performance of these models when multiple targets are available. Our results suggest that GBDTs perform well across all targets, and that multitask learning does not provide a consistent improvement. |
| first_indexed | 2024-04-10T06:04:46Z |
| format | Article |
| id | doaj.art-642d6fad0d344f849ee0d52f917d2016 |
| institution | Directory Open Access Journal |
| issn | 2352-9148 |
| language | English |
| last_indexed | 2024-04-10T06:04:46Z |
| publishDate | 2023-01-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Informatics in Medicine Unlocked |
| spelling | doaj.art-642d6fad0d344f849ee0d52f917d20162023-03-03T04:24:49ZengElsevierInformatics in Medicine Unlocked2352-91482023-01-0137101161High-performance prediction models for prostate cancer radiomicsLars Johannes Isaksson0Marco Repetto1Paul Eugene Summers2Matteo Pepa3Mattia Zaffaroni4Maria Giulia Vincini5Giulia Corrao6Giovanni Carlo Mazzola7Marco Rotondi8Federica Bellerba9Sara Raimondi10Zaharudin Haron11Sarah Alessi12Paula Pricolo13Francesco Alessandro Mistretta14Stefano Luzzago15Federica Cattani16Gennaro Musi17Ottavio De Cobelli18Marta Cremonesi19Roberto Orecchia20Davide La Torre21Giulia Marvaso22Giuseppe Petralia23Barbara Alicja Jereczek-Fossa24Division of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, Italy; Department of Oncology and Hemato-Oncology, University of Milan, Milan, Italy; Corresponding author at: Division of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, Italy.Digital Industries, Siemens Italy, Italy; Department of Economics, Management and Statistics, University of Milan-Bicocca, Italy; SKEMA Business School, Université Côte d’Azur, Sophia Antipolis Campus, FranceDivision of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, ItalyDivision of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, ItalyDivision of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, Italy; Corresponding author.Division of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, ItalyDivision of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, ItalyDivision of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, Italy; Department of Oncology and Hemato-Oncology, University of Milan, Milan, ItalyDivision of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, Italy; Department of Oncology and Hemato-Oncology, University of Milan, Milan, ItalyDepartment of Experimental Oncology, IEO European Institute of Oncology IRCCS, Milan, ItalyDepartment of Experimental Oncology, IEO European Institute of Oncology IRCCS, Milan, ItalyRadiology Department, National Cancer Institute, Putrajaya, MalaysiaDivision of Radiology, IEO European Institute of Oncology IRCCS, Milan, ItalyDivision of Radiology, IEO European Institute of Oncology IRCCS, Milan, ItalyDivision of Urology, IEO European Institute of Oncology IRCCS, Milan, ItalyDivision of Urology, IEO European Institute of Oncology IRCCS, Milan, ItalyMedical Physics Unit, IEO European Institute of Oncology IRCCS, Milan, ItalyDepartment of Oncology and Hemato-Oncology, University of Milan, Milan, Italy; Division of Urology, IEO European Institute of Oncology IRCCS, Milan, ItalyDepartment of Oncology and Hemato-Oncology, University of Milan, Milan, Italy; Division of Urology, IEO European Institute of Oncology IRCCS, Milan, ItalyRadiation Research Unit, IEO European Institute of Oncology IRCCS, Milan, ItalyScientific Directorate, IEO European Institute of Oncology IRCCS, Milan, ItalyDepartment of Oncology and Hemato-Oncology, University of Milan, Milan, Italy; SKEMA Business School, Université Côte d’Azur, Sophia Antipolis Campus, FranceDivision of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, Italy; Department of Oncology and Hemato-Oncology, University of Milan, Milan, ItalyDepartment of Oncology and Hemato-Oncology, University of Milan, Milan, Italy; Precision Imaging and Research Unit, Department of Medical Imaging and Radiation Sciences, IEO European Institute of Oncology IRCCS, Milan, ItalyDivision of Radiation Oncology, IEO European Institute of Oncology IRCCS, Milan, Italy; Department of Oncology and Hemato-Oncology, University of Milan, Milan, ItalyWhen researchers are faced with building machine learning (ML) radiomic models, the first choice they have to make is what model to use. Naturally, the goal is to use the model with the best performance. But what is the best model? It is well known in ML that modern techniques such as gradient boosting and deep learning have better capacity than traditional models to solve complex problems in high dimensions. Despite this, most radiomics researchers still do not focus on these models in their research. As access to high-quality and large data sets increase, these high-capacity ML models may become even more relevant. In this article, we use a large dataset of 949 prostate cancer patients to compare the performance of a few of the most promising ML models for tabular data: gradient-boosted decision trees (GBDTs), multilayer perceptions, convolutional neural networks, and transformers. To this end, we predict nine different prostate cancer pathology outcomes of clinical interest. Our goal is to give a rough overview of how these models compare against one another in a typical radiomics setting. We also investigate if multitask learning improves the performance of these models when multiple targets are available. Our results suggest that GBDTs perform well across all targets, and that multitask learning does not provide a consistent improvement.http://www.sciencedirect.com/science/article/pii/S2352914823000035RadiomicsProstate cancerDeep learningGradient boost |
| spellingShingle | Lars Johannes Isaksson Marco Repetto Paul Eugene Summers Matteo Pepa Mattia Zaffaroni Maria Giulia Vincini Giulia Corrao Giovanni Carlo Mazzola Marco Rotondi Federica Bellerba Sara Raimondi Zaharudin Haron Sarah Alessi Paula Pricolo Francesco Alessandro Mistretta Stefano Luzzago Federica Cattani Gennaro Musi Ottavio De Cobelli Marta Cremonesi Roberto Orecchia Davide La Torre Giulia Marvaso Giuseppe Petralia Barbara Alicja Jereczek-Fossa High-performance prediction models for prostate cancer radiomics Informatics in Medicine Unlocked Radiomics Prostate cancer Deep learning Gradient boost |
| title | High-performance prediction models for prostate cancer radiomics |
| title_full | High-performance prediction models for prostate cancer radiomics |
| title_fullStr | High-performance prediction models for prostate cancer radiomics |
| title_full_unstemmed | High-performance prediction models for prostate cancer radiomics |
| title_short | High-performance prediction models for prostate cancer radiomics |
| title_sort | high performance prediction models for prostate cancer radiomics |
| topic | Radiomics Prostate cancer Deep learning Gradient boost |
| url | http://www.sciencedirect.com/science/article/pii/S2352914823000035 |
| work_keys_str_mv | AT larsjohannesisaksson highperformancepredictionmodelsforprostatecancerradiomics AT marcorepetto highperformancepredictionmodelsforprostatecancerradiomics AT pauleugenesummers highperformancepredictionmodelsforprostatecancerradiomics AT matteopepa highperformancepredictionmodelsforprostatecancerradiomics AT mattiazaffaroni highperformancepredictionmodelsforprostatecancerradiomics AT mariagiuliavincini highperformancepredictionmodelsforprostatecancerradiomics AT giuliacorrao highperformancepredictionmodelsforprostatecancerradiomics AT giovannicarlomazzola highperformancepredictionmodelsforprostatecancerradiomics AT marcorotondi highperformancepredictionmodelsforprostatecancerradiomics AT federicabellerba highperformancepredictionmodelsforprostatecancerradiomics AT sararaimondi highperformancepredictionmodelsforprostatecancerradiomics AT zaharudinharon highperformancepredictionmodelsforprostatecancerradiomics AT sarahalessi highperformancepredictionmodelsforprostatecancerradiomics AT paulapricolo highperformancepredictionmodelsforprostatecancerradiomics AT francescoalessandromistretta highperformancepredictionmodelsforprostatecancerradiomics AT stefanoluzzago highperformancepredictionmodelsforprostatecancerradiomics AT federicacattani highperformancepredictionmodelsforprostatecancerradiomics AT gennaromusi highperformancepredictionmodelsforprostatecancerradiomics AT ottaviodecobelli highperformancepredictionmodelsforprostatecancerradiomics AT martacremonesi highperformancepredictionmodelsforprostatecancerradiomics AT robertoorecchia highperformancepredictionmodelsforprostatecancerradiomics AT davidelatorre highperformancepredictionmodelsforprostatecancerradiomics AT giuliamarvaso highperformancepredictionmodelsforprostatecancerradiomics AT giuseppepetralia highperformancepredictionmodelsforprostatecancerradiomics AT barbaraalicjajereczekfossa highperformancepredictionmodelsforprostatecancerradiomics |