Table 1 Overview of studies included in the systematic review.
Fusion strategy | Year | Author | Clinical ___domain | Outcome | Fusion details | Input: medical imaging | Input: non-imaging data | Number of samples | Model performance |
|---|---|---|---|---|---|---|---|---|---|
Early | 2017 | Thung et al.25 | Neurology | Diagnosis of Alzheimer’s disease | Software or manually extracted features | PET, MRI | Patient data (age, sex, education) Genetic data (APOE4) | 805 | Fusion: 63.6% Accuracy MRI + PET: 61.1% Accuracy MRI: 58.0% Accuracy |
Early | 2018 | An et al.35 | Ophthalmology | Glaucoma classification | Software or manually extracted features | Optical coherence tomography, Laser speckle flowgraphy | Patient data (age, sex, spherical equivalent) | 163 | Fusion: 87.8% Accuracy |
Early (multi-stage) | 2018 | Bhagwat et al.33 | Neurology | Prediction of clinical symptom trajectories in Alzheimer’s disease | Software or manually extracted features | MRI | Patient data (age, clinical scores) Genetic data (APOE4) | 1302 | Fusion: 0.99 AUROC MRI: 0.83 AUROC Clinical Attributes: 0.97 AUROC |
Early | 2018 | Kharazmi et al.26 | Dermatology | Basal cell carcinoma detection | CNN extracted features | Dermoscopic images | Patient data (age, sex, elevation, lesion ___location, lesion size) | 1191 | Fusion: 91.1% Accuracy Dermoscopic Images: 84.7% Accuracy Patient Profile: 75.6% Accuracy |
Early | 2018 | Liu et al.34 | Neurology | Prediction of rupture risk in anterior communicating artery aneurysms | Software or manually extracted features | CT | Patient data (age, sex, hypertension, smoking habits) | 594 | Fusion: 0.928 AUROC |
Early | 2018 | Liu et al.30 | Radiology | Bone age assessment | CNN extracted features | X-ray | Patient data (age, sex) | 11,858 | Fusion: 0.455 Mean Absolute Error |
Early | 2018 | Yap et al.27 | Dermatology | Classification of skin lesion and detection of melanoma | CNN extracted features | Macroscopic images, Dermoscopic images | Patient data (age, sex, body ___location) | 2917 | Fusion: 0.888 AUROC Macroscopic + Dermoscopic Images: 0.888 AUROC Patient metadata: 0.810 AUROC |
Early | 2019 | Hyun et al.32 | Radiology/Oncology | Lung cancer | Software or manually extracted features | PET/CT | Patient data (age, sex, tumor size, smoking status) | 396 | Fusion: 0.859 AUROC |
Early | 2019 | Li et al.28 | Neurology | Prediction of Alzheimer’s disease | CNN extracted features | MRI | Assessments (Alzheimer’s Disease Assessment Scale-Cognitive subscale, Rey Auditory Verbal Learning Test, Functional Assessment Questionnaire, and Mini-Mental State Examination) Patient data (age, gender, education, APOE4) | 822 | Fusion: 0.901 C-index Cognitive data: 0.896 C-index |
Early | 2019 | Nie et al.31 | Radiology/Oncology | Prediction of survival time for brain tumor patients | CNN extracted features | MRI | Patient data (age, tumor size, histological type) | 93 | Fusion: 90.66% Accuracy MRI: 81.04% Accuracy Demographics and tumor features: 62.96% Accuracy |
Early | 2020 | Purwar et al.29 | Hematology | Detection of microcytic hypochromia | CNN extracted features | Images of Red Blood Cells | Blood test (complete blood count, haematocrit, HCT, MCV, MCH, MCHC, RDW, hemoglobin A1 hemoglobin A2, hemoglobin F, Mentzer index) | 20 | Fusion: 1.00 AUROC Images: 0.88 AUROC Blood count features: 0.93 AUROC |
Joint* | 2018 | Kawahara et al.39 | Dermatology | Melanoma classification and binary classification of each of the seven-point checklist for melanoma | Multimodal multi-task | Dermoscopic Images, Clinical Images | Patient data (sex, lesion ___location) | 1011 | Fusion: 73.7% Accuracy Dermoscopic Images: 72.5% Accuracy Clinical Images: 64.1% Accuracy |
Joint | 2018 | Spasov et al.36 | Neurology | Prediction of Alzheimer’s disease | – | MRI | Patient data (age, sex, race, education, biofluids) Genetic data (APOE4) Assessments (Clinical Dementia rating [CSRSB], Alzheimer’s disease assessment scale, Rey auditory verbal learning test) | 376 | Fusion: 1.00 AUROC |
Joint | 2019 | Yala et al.37 | Radiology/Oncology | Breast cancer risk prediction | – | Mammograms | Patent data (age, weight, height, menarche age, menopausal status, detailed family history of breast and ovarian cancer, BRCA mutation status, history of atypical hyperplasia, history of lobular carcinoma in situ, and breast density) | 88,994 | Fusion: 0.70 AUROC Mammograms: 0.68 AUROC Risk scores: 0.67 AUROC |
Joint Late | 2019 | Yoo et al.38 | Neurology | Predicting status conversion to multiple sclerosis within two years | CNN extracted feature Averaging (Late) | MRI | Patient data (sex, extended disability status scale, uni- vs. multifocal clinically isolated syndrome (CIS) at onset, ___location of initial CIS event) | 140 | Joint Fusion: 0.746 AUROC Late Fusion: 0.724 AUROC MRI: 0.718 AUROC Clinical Features: 0.679 AUROC |
Late | 2018 | Qiu et al.41 | Neurology | Mild cognitive impairment (MCI) | Majority voting | MRI | Assessments (Mini Mental State Examination (MMSE), Wechsler Memory Scale Logical Memory (LM) test) | 386 | Fusion: 90.9% Accuracy MRI: 83.1% Accuracy MMSE: 84.3% Accuracy LM: 89.1% Accuracy |
Late | 2018 | Reda et al.40 | Radiology/Oncology | Prostate cancer diagnosis | Meta classifier | MRI | PSA blood test | 18 | Fusion: 94.4% Accuracy MRI: 88.89% Accuracy PSA: 77.78% Accuracy |
