AI-based classification of three common malignant tumors in neuro-oncology: A multi-institutional comparison of machine learning and deep learning methods - 09/09/23

Doi : 10.1016/j.neurad.2023.08.007

Girish Bathla ^a,^b,^⁎ , Durjoy Deb Dhruba ^c, Neetu Soni ^a,^d, Yanan Liu ^e, Nicholas B Larson ^f, Blake A Kassmeyer ^f, Suyash Mohan ^g, Douglas Roberts-Wolfe ^g, Saima Rathore ^h, Nam H Le ^c, Honghai Zhang ^c, Milan Sonka ^c, Sarv Priya ^a
^a Department of Radiology, University of Iowa Hospitals and Clinics, 200 Hawkins Drive, Iowa City, IA 52242, USA
^b Department of Radiology, Mayo Clinic, 200 1st Street SW, Rochester, MN 55902, USA
^c Electrical and Computer Engineering, University of Iowa, 4016 Seamans Center for the Engineering Arts and Sciences, Iowa City, IA 52242 USA
^d Department of Imaging Sciences, University of Rochester Medical Center, 601 Elmwood Ave, Box 648, Rochester, NY 14642, USA
^e Advanced Pulmonary Physiomic Imaging Laboratory (APPIL), University of Iowa, 200 Hawkins Drive, Iowa City, IA, 52242 USA
^f Division of Clinical Trials and Biostatistics, Department of Quantitative Health Sciences, Mayo Clinic, 200 1st Street SW, Rochester, MN 55902, USA
^g Department of Radiology, Perelman School of Medicine, University of Pennsylvania, 3400 Spruce Street, Philadelphia, PA 19104 USA
^h Senior research scientist, Avid Radiopharmaceuticals, 3711 Market Street, Philadelphia, PA 19104, USA

^⁎Corresponding author at: Department of Radiology, Mayo Clinic, 200, 1st St SW Rochester, MN, USA.Department of RadiologyMayo Clinic, 200, 1st St SW RochesterMNUSA

En prensa. Pruebas corregidas por el autor. Disponible en línea desde el Saturday 09 September 2023
This article has been published in an issue click here to access

Highlights

•	We compared ML and DL pipelines for a three-class tumor classification.
•	Both ML and DL pipelines showed similar robust performance.
•	Features were derived from both the tumor and surrounding tissues for top models.
•	T1-CE sequence may be the most useful sequence overall.

El texto completo de este artículo está disponible en PDF.

Abstract

Purpose

To determine if machine learning (ML) or deep learning (DL) pipelines perform better in AI-based three-class classification of glioblastoma (GBM), intracranial metastatic disease (IMD) and primary CNS lymphoma (PCNSL).

Methodology

Retrospective analysis included 502 cases for training (208 GBM, 67 PCNSL and 227 IMD), with external validation on 86 cases (27:27:32). Multiparametric MRI images (T1W, T2W, FLAIR, DWI and T1-CE) were co-registered, resampled, denoised and intensity normalized, followed by semiautomatic 3D segmentation of the enhancing tumor (ET) and peritumoral region (PTR). Model performance was assessed using several ML pipelines and 3D-convolutional neural networks (3D-CNN) using sequence specific masks, as well as combination of masks. All pipelines were trained and evaluated with 5-fold nested cross-validation on internal data followed by external validation using multi-class AUC.

Results

Two ML models achieved similar performance on test set, one using T2-ET and T2-PTR masks (AUC: 0.885, 95% CI: [0.816, 0.935] and another using T1-CE-ET and FLAIR-PTR mask (AUC: 0.878, CI: [0.804, 0.930]). The best performing DL models achieved an AUC of 0.854, (CI [0.774, 0.914]) on external data using T1-CE-ET and T2-PTR masks, followed by model derived from T1-CE-ET, ADC-ET and FLAIR-PTR masks (AUC: 0.851, CI [0.772, 0.909]).

Conclusion

Both ML and DL derived pipelines achieved similar performance. T1-CE mask was used in three of the top four overall models. Additionally, all four models had some mask derived from PTR, either T2WI or FLAIR.

El texto completo de este artículo está disponible en PDF.

Keywords : Artificial intelligence, Radiomics, Glioblastoma, Metastasis, CNS lymphoma