Table 1. Studies using deep learning in hematology whole slide imaging interpretation.
| Biopsy sample | Clinical Task | Training size | DL model: Patch Feature Extractor | DL model: Patch Feature Aggregator | Testing dataset | Testing results | References |
|---|---|---|---|---|---|---|---|
| LN | Differentiate DLBCL, BL, SLL, and benign | 128 | CNN on manually cropped area | None | Internal | Accuracy 95% | Achi et al., 2019107 |
| LN | Differentiate DLBCL from various benign and malignant LN samples | 1,754 | Majority-voting of 17 CNNs on manually cropped area | None | External | Accuracy >99% | Li et al., 2020108 |
| LN | Differentiate DLBCL, FL, and benign | 388 | CNN on manually cropped area | None | Internal | Accuracy 90% AUC 0.95 |
Miyoshi et al., 2020109 |
| LN | Differentiate DLBCL, SLL, and benign | 629 | CNN on manually cropped area | None | External | Accuracy 96% | Steinbuss et al., 2021110 |
| LN and other biopsy sites | Predict MYC rearrangement on H&E stained DLBCL WSIs | 287 | CNN | Not clearly specified | External | Accuracy 74% AUC 0.83 |
Swiderska-Chadaj et al., 2021111 |
| LN | Differentiate FL and benign hyperplasia | 378 | CNN | Mean pooling | External | AUC 0.66 | Syrykh et al., 2020112 |
| Skin | Annotate CD30+ regions on CD30-stained WSIs to diagnose CD30+ LPD | 28 | CNN | Local self-attention, sum pooling | Internal | Accuracy 96% AUC 0.99 |
Zheng et al., 2023113 |
| BM | Predict mutations on H&E stained MDS WSIs | 236 | Pretrained CNN | Mean pooling | Internal | AUC varies on mutations, as high as 0.94 | Bruck et al., 2021114 |
| BM | Differentiate AML, CML, ALL, CLL, and MM | 129 | Pretrained CNN | Attention | External | Accuracy 94% AUC 0.97 |
Wang et al., 2022115 |
| BM | Differentiate ET and prePMF | 226 | Pretrained CNN | Attention | Internal | Accuracy 92% AUC 0.90 |
Srisuwananukorn et al., 2023116 |
| BM | Differentiate AL, MM, LPD, and normal | 556 | Pretrained YOLO for cell detection and feature extraction | Attention | Internal | Average F1 score 0.57 | Mu et al, 2023117 |
DL, deep learning. LN, lymph nodes. DLBCL, diffuse large B-cell lymphoma. BL, Burkitt’s lymphoma. SLL, small lymphocytic lymphoma. CNN, convolutional neural network. FL, follicular lymphoma. AUC, area under curve. WSI, whole slide image. LPD, lymphoproliferative disease. BM, bone marrow. MDS, myelodysplastic syndrome. AML, acute myeloid leukemia. CML, chronic myeloid leukemia. ALL, acute lymphoblastic leukemia. CLL, chronic lymphocytic leukemia. MM, multiple myeloma. ET, essential thrombocythemia. prePMF, prefibrotic primary myelofibrosis. AL, acute leukemia.