Application of Morphogo based on convolutional neural network for morphological identification of bone marrow nucleated cells

doi:10.21203/rs.3.rs-4126940/v1

Introduction:

The morphological examination of bone marrow (BM) cells, commonly used for diagnosing hematological diseases, heavily relies on the expertise of pathologists. This approach is time-consuming, labor-intensive, subjective, and lacks objectivity. Therefore, it is crucial to develop automated analysis systems to aid in the diagnosis of hematological diseases.

Methods

The BM smears from patients with hematological diseases were collected from Dian Diagnostics between September 2021 and December 2021. These smears were classified into five groups based on varying degrees of cell morphological alterations. Images of the BM nucleated cells were captured using the Morphogo system, and its performance in cell identification was compared with that of pathologists.

Results

The Morphogo system demonstrated a high performance in identifying BM nucleated cells, with a sensitivity of 0.9362, specificity of 0.9977, PPV of 0.8354, NPV of 0.9974, and accuracy of 0.9954. Comparison between the percentage of BM nucleated cells identified by the Morphogo system and pathologists showed almost perfect agreement, with an average Kappa value of 0.8695 for 25 cell classes. The practical utility of the Morphogo system was evaluated in hematological diseases, with pathologists achieving averaged sensitivity, specificity, PPV, NPV and accuracy ranging from 0.9098 to 0.9868 when using the system for disease diagnosis. The diagnostic results were consistent with those made by pathologists using a microscope, with an average Kappa value of 0.9096.

Conclusion

Morphogo system had the potential to assist pathologists in diagnosis of hematological diseases by improving the efficiency of identification of BM nucleated cells.

Morphogo system

Bone marrow nucleated cells

Morphology

Hematological diseases

The morphological examination of nucleated cells in bone marrow (BM) is a critical diagnostic procedure for hematological diseases, heavily reliant on the expertise of pathologists. However, this process is time-consuming and labor-intensive, potentially leading to delays in treatment initiation for patients [1–9]. Therefore, there is an urgent need for new technology to assist with this manual inspection.

Convolutional neural network (CNN) is a type of feedforward neural network that incorporates convolution computation and depth structure [10–11]. As a deep learning algorithm, CNN has demonstrated outstanding performance in image recognition and feature extraction due to its unique hierarchical structure and parameter learning ability [8, 12]. In clinicopathological examination, pathological image analysis plays a crucial role in disease diagnosis, and the utilization of CNN significantly enhances the accuracy and efficiency of this process [13–20]. However, due to the complexity of BM cell morphology, CNN is seldom utilized for nucleated cell recognition in BM.

In this study, we collected BM smears from 207 patients with different blood diseases, developed the CNN-based Morphogo system, and demonstrated its effectiveness in the identification of BM nucleated cells and its clinical application value in assisting the diagnosis of hematologic diseases.

2.1 Sources and Classification of Sample

This was a retrospective study in which 207 BM cases were collected from Dian Diagnostics between September 2021 and December 2021. The smears were categorized into five groups (G1 to G5) based on the degree of pathological and cell morphological changes, as recommended by pathologists. The diseases included in each group are as follows: G1 - acute myeloid leukemia (AML), acute lymphoblastic leukemia (ALL), myelodysplastic syndrome transformation into acute leukemia (MDS-AL), chronic myeloid leukemia diagnosed in accelerated phase (CML-AP), and lymphoma; G2 - megaloblastic anemia (MA), plasma cell myeloma (PCM), myelodysplastic syndrome (MDS), chronic myeloid leukemia (CML), and myeloproliferative neoplasms (MPN); G3 - immune thrombocytopenia (ITP), chronic lymphoblastic leukemia (CLL) and other diseases with slight changes in cell morphology; G4 - iron deficiency anemia (IDA), secondary anemia (SA), secondary leukaemia (SL), and aplastic anaemia (AA); and G5 - BM dilution cases. All BM smears were well stained using the Wright-Giemsa method, and the quality of both the smear and staining met the recommendations outlined in the nation guide to clinical laboratory procedures (NGCLP, fourth edition) or by the international council for standardization in hematology (ICSH) [8, 21]. The study received approval from the institutional review committee of all participating agencies. Detailed information on enrolled BM cases can be found in Table 1.

Table 1

Information statistics table of BM smears.
Class of morphology	Sample	Number
G1	MDS-AL	10
	AML	16
	ALL	5
	CML-AP	2
	Lymphoma	2
G2	MA	9
	PCM	11
	MDS	13
	CML	4
	MPN	5
G3	ITP	7
	CLL	17
	Others	18
G4	IDA	18
	SA	7
	SL	5
	AA	2
	Others	7
G5	Dilution cases	49

MDS-AL: myelodysplastic syndrome transformation into acute leukemia; AML: acute myeloid leukemia; ALL: acute lymphoblastic leukemia; CML-AP: chronic myeloid leukemia diagnosed in accelerated phase; MA: megaloblastic anemia; PCM: plasma cell myeloma; MDS: myelodysplastic syndrome; CML: chronic myeloid leukemia; MPN: myeloproliferative neoplasms; ITP: immune thrombocytopenia; CLL: chronic lymphoblastic leukemia; IDA: iron deficiency anemia; SA: secondary anemia; SL: secondary leukemia; AA: aplastic anemia.

2.2 CNN construction

CNN consists of three parts: convolution layer, pooling layer and fully connected layer. The filter of the convolution layer alters or enhances cell images by strengthening or removing image features such as blurring, sharpening, embossing, and edge detection. Additionally, the pooling layer reduces data dimensionality, space size of image features, calculation amount and parameters in the neural network by combining output from neural clusters into a single neuron in the next layer. This also improves stability by controlling overfitting. After the system completes the work of the convolution layer and pooling layer, the fully connected layer is formatted. Then, fully connected layer extracts the advanced image features generated by the convolution layer and pooling layer and classifies the input cell images. Finally, the results of cell images classification are output as vector values. Each value represents the classification probability under different elements [14, 22–24].

2.3 Data digitization

Morphogo system is a CNN-based artificial Intelligence (AI) system developed by Hangzhou Zhiwei Information and Technology Ltd that is used to perform differential count of BM nucleated cells automatically. After training with more than 2.8 million BM nucleated cells, Morphogo system can identify over 35 types of BM nucleated cells. The acquisition process of digital slides is as follows: (1) the whole slide imaging (WSI) is obtained by scanning the BM smears with a 40X objective lens; (2) the 100X objective lens is used to obtain detailed cell images in the region of interest (ROI), and then AI algorithms locate, segment and identify BM nucleated cells on the digital images; (3) the cell differential count results are automatically calculated for each slide. Morphogo system can classify and count up to 9999 BM nucleated cells per smear, which can satisfy the needs of laboratory work, research and education at the same time.

2.4 Morphogo system workflow

The workflow by which pathologists and Morphogo system identify BM nucleated cells was illustrated in Fig. 1. In a manual morphology examination, pathologists classified and counted 200–500 nucleated cells on every BM smear, then issued a cytology report based on morphology information as well as other clinical information such as CBC and medical history. In an automated morphology assessment using Morphogo system, the system firstly scanned the BM smear to obtain a high-quality digital smear and BM cell images in ROI, then the CNN embedded in Morphogo system was used to locate, segment and identify BM cells in the images. The acquisition process was fully automated. Then, more than three pathologists reviewed the digital BM smears and proofread the results of cell pre-classification based on AI on a computer monitor. If cells were mistakenly identified by Morphogo system and they would be reclassified by pathologists. Finally, Morphogo system automatically calculated the proportion of each BM cell type and the degree of hyperplasia of each BM smear, and then generated a morphology report.

2.5 Morphogo system evaluation

To evaluate the cell classification performance of Morphogo system in different hematological disease, the BM nucleated cells were annotated into 25 categories: proerythroblast, early erythroblast, intermediate erythroblast, later erythroblast, myeloblast, promyelocyte, neutrophilic myelocyte, neutrophilic metamyelocyte, band neutrophil, segmented neutrophil, eosinophilic myelocyte, eosinophilic metamyelocyte, band eosinophil, segmented eosinophil, basophil, monoblast, promonocyte, monocyte, lymphoblast, prolymphocyte, mature lymphocyte, plasmablast, immature plasma, plasma cell and others including smudge cell, histocyte and mast cell according to WHO classification. Cell classification performance was evaluated in terms of sensitively, specificity, positive predict value (PPV), negative predict value (NPV) and accuracy [25–27]. Meanwhile, we collected the results of manual microscope examination of all BM smears and other clinical examination results of all cases and compared the kappa values for evaluating the agreement of pathologists using Morphogo system and manual microscope in diseases diagnosis.

2.6 Statistical analysis

Excel version 2016 was utilized for the analysis of sensitivity, specificity, PPV, NPV and accuracy of Morphogo’s cell classification, assuming pathologists’ annotations as the gold standard. The correlations of cell proportions were visualized using GraphPad Prism 7. Kappa was calculated using SPSS 20 to assess consistency [28–29].

In this study, the calculation formulas were as follows (TP: true positive; TN: true negative; FP: false positive; FN: false negative):

$$Specificity=\frac{\text{T}N}{TN+FP}*100\%$$

$$PPV=\frac{\text{T}P}{\text{T}P+FP}*100\%$$

$$NPV=\frac{\text{T}N}{\text{F}N+TN}*100\%$$

$$Accuracy=\frac{\text{T}P+TN}{\text{T}P+FP+TN+FN}*100\%$$

3.1 The Morphogo system is capable of accurately identifying BM nucleated cells

The 68,610 high-resolution digital images of BM nucleated cells captured by the Morphogo system were segmented into 25 groups and annotated. To assess the performance of the Morphogo system in classifying cells across various pathological conditions, it was applied to a dataset encompassing 19 types of hematological diseases in patient cases. The evaluation metrics for each disease condition are summarized in Table 2.

The Morphogo system demonstrated a sensitivity exceeding 0.7995 in classifying BM nucleated cells, with an average sensitivity of 0.9362. The PPV varied significantly among different cell types, ranging from 0.3698 to 0.9991, and exceeding 0.95 for promyelocyte, neutrophil, erythroid cells, mature lymphocyte, and other cell types. Conversely, the PPV for eosinophil, lymphocyte, prolymphocyte, monoblast, promonocyte, plasmablast and immature plasma was below 0.80. The NPV remained consistently high at a range of 0.9745 to 1.0000 across all classes of BM nucleated cells with eosinophilic myelocyte and monoblast exhibiting an NPV of 1.0000. The accuracy of the Morphogo system in classifying BM nucleated cells was remarkably high at a range from 0.9713 to 0.9999 with an average accuracy of 0.9954.

Table 2

Statistics of Morphogo system to classify BM nucleated cells.
Class of cells	Sensitivity	Specificity	PPV	NPV	Accuracy
Myeloblast	0.9533	0.9960	0.8542	0.9989	0.9950
Promyelocyte	0.9689	0.9998	0.9851	0.9995	0.9993
Neutrophilic myelocyte	0.9509	0.9995	0.9915	0.9971	0.9968
Neutrophilic metamyelocyte	0.9263	0.9997	0.9953	0.9945	0.9945
Band neutrophil	0.9787	0.9998	0.9978	0.9979	0.9979
Segmented neutrophil	0.9742	0.9999	0.9991	0.9960	0.9964
Eosinophilic myelocyte	0.9880	0.9988	0.7440	1.0000	0.9987
Eosinophilic metamyelocyte	0.9950	0.9949	0.7428	0.9999	0.9949
Band eosinophil	0.8814	0.9998	0.7536	0.9999	0.9997
Segmented eosinophil	0.9447	0.9972	0.7341	0.9995	0.9967
Basophil	0.8921	0.9994	0.8851	0.9994	0.9988
Proerythroblast	0.8791	0.9999	0.9524	0.9998	0.9998
Early erythroblast	0.8963	0.9998	0.9629	0.9994	0.9992
Intermediate erythroblast	0.9708	0.9976	0.9760	0.9970	0.9951
Later erythroblast	0.9755	0.9991	0.9898	0.9978	0.9971
Lymphoblast	0.9848	0.9937	0.6015	0.9999	0.9936
Prolymphocyte	0.7995	0.9917	0.3698	0.9988	0.9906
Mature lymphocyte	0.8373	0.9996	0.9975	0.9668	0.9713
Monoblast	0.9972	0.9978	0.7084	1.0000	0.9978
Promonocyte	0.8735	0.9959	0.5648	0.9992	0.9951
Monocyte	0.8257	0.9976	0.8860	0.9961	0.9938
Plasmablast	1.0000	0.9999	0.6316	1.0000	0.9999
Immature plasma	0.9980	0.9866	0.6923	0.9999	0.9869
Plasma cell	0.9137	0.9981	0.8729	0.9987	0.9969
Others	0.9992	0.9995	0.9967	0.9999	0.9995

3.2 The Morphogo system and the pathologist are extremely consistent in identifying BM nucleated cells

The Morphogo system was subjected to correlation and consistency analysis in order to enhance our understanding of its concordance with pathologists in the classification and enumeration of BM nucleated cells.

Table 3

Evaluation of the consistence of BM nucleated cells between Morphogo system pre-classification and manual proofreading in terms of Cohen kappa coefficient.
Class of cells	Kappa	P value
Myeloblast	0.8980	0.0000
Promyelocyte	0.9770	0.0000
Neutrophilic myelocyte	0.9690	0.0000
Neutrophilic metamyelocyte	0.9310	0.0000
Band neutrophil	0.9870	0.0000
Segmented neutrophil	0.9840	0.0000
Eosinophilic myelocyte	0.8480	0.0000
Eosinophilic metamyelocyte	0.8480	0.0000
Band eosinophil	0.8120	0.0000
Segmented eosinophil	0.8250	0.0000
Basophil	0.8880	0.0000
Proerythroblast	0.9140	0.0000
Early erythroblast	0.9280	0.0000
Intermediate erythroblast	0.9710	0.0000
Later erythroblast	0.9810	0.0000
Lymphoblast	0.7440	0.0000
Prolymphocyte	0.5020	0.0000
Mature lymphocyte	0.8930	0.0000
Monoblast	0.8270	0.0000
Promonocyte	0.6840	0.0000
Monocyte	0.8520	0.0000
Plasmablast	0.7740	0.0000
Immature plasma	0.8110	0.0000
Plasma cell	0.8910	0.0000
Others	0.9980	0.0000

The results revealed a robust positive correlation between the Morphogo system and pathologists in classifying 25 BM cell types, with a P-value of less than 0.001 and a mean Kappa value of 0.8695 (Fig. 2, Table 3). It is worth noting that the Morphogo system exhibits limited capability in identifying prolymphocytes, as indicated by a Kappa value of only 0.5020 (Table 3). Overall, the findings from both correlation and consistency analyses strongly support substantial agreement between the Morphogo system and pathologists in accurately identifying BM nucleated cells.

3.3 The application of Morphogo system can effectively diagnose hematological system diseases.

To further validate the diagnostic utility of the Morphogo system in hematological diseases, we compared the morphological diagnoses made by pathologists utilizing the Morphogo system with the original cytology diagnoses, treating the latter as the gold standard.

Table 4

Statistics of Morphogo system to hematological diseases diagnosis with the changes of cells morphologies.
Class of morphology	Sensitivity	Specificity	PPV	NPV	Accuracy
G1	0.8857	1.0000	1.0000	0.9773	0.9807
G2	0.8810	0.9939	0.9737	0.9704	0.9710
G3	0.9048	0.9697	0.8837	0.9756	0.9565
G4	0.8974	0.9702	0.8750	0.9760	0.9565
G5	0.9800	1.0000	1.0000	0.9937	1.0000

Based on the data in Table 4, the Morphogo system exhibited sensitivities ranging from 0.8810 to 0.8974 in G1, G2, and G4 groups, slightly lower than those in G3 and G5. It is worth noting that the system demonstrated high specificities exceeding 0.95 across all groups, reaching a peak of 1.0000 in G1 and G5. The PPV in G1 and G5 was perfect at 1.0000, with an average of 0.9465. The NPVs consistently exceeded 0.95, with a value of 0.9937 reached in group G5 specifically. Furthermore, the accuracy of the Morphogo system was also high, surpassing 0.95 in all groups and peaking at a perfect score of 1.0000 in group G5. Additionally, Table 5 presents the results of the concordance analysis between the Morphogo system and microscopy, further confirming its effectiveness in diagnosing hematological diseases.

Table 5

Evaluation of the consistence of hematological diseases between Morphogo system and microscope in terms of Cohen kappa coefficient.
Class of morphology	Kappa	P value
G1	0.9280	0.0000
G2	0.9070	0.0000
G3	0.8670	0.0000
G4	0.8590	0.0000
G5	1.0000	0.0000

CNN model trained by large datasets has been proved to be able to achieve human-level ability in identification and classification of various medical images, such as detecting the classification of diabetic retinopathy[8, 27, 30]. However, there is not an AI-based device that is currently in use of BM morphology assessment in clinical laboratories.

The morphology of BM nucleated cells varied in different pathological conditions, thus the Morphogo system’s analytical performance was evaluated in 19 types of hematological diseases. Due to small numbers of cases in certain pathological conditions, the BM smears were divided into five groups (G1-G5) according to degrees of cell morphology changes. In this study, the G1 was made up of leukemia with a significantly increased of blasts in the BM including MDS-AL, AML, ALL, CML-AP and malignant lymphoma [31–33]. The G2 consisted of MA, PCM, MDS, CML and MPN. Cells in these diseases has moderate or obvious morphological changes. MA is characterized by the presence of megaloblast in BM due to mature asynchronously between nucleus and cytoplasm [34]. Plasma cells in PCM are highly pleomorphic, such as diffuse sheet growth pattern and immature cell morphology[2]. MDS or MPN is usually characterized by dysplasia in one or more myeloid lineages in BM [35, 36]. CML is a kind of cancer with an uncontrolled increased in the number of myeloid cells [37]. Hematological diseases in G3 were accompanied by mild changes in BM nucleated cell morphology, including ITP and CLL. ITP is an acquired autoimmune disease with increased platelet destruction and decrease platelet production [38]. B lymphocytes debris is a morphology feature of CLL [39]. IDA, SA, SL, and AA were classified into G4, and these diseases are basically not accompanied by cell morphological changes [40–42]. The G5 was dilution cases. The diagnostic results that pathologists made based on Morphogo system’s morphological findings were compared with the original cytology diagnoses based on manual microscopic examination. Morphogo system displayed a good ability to assist diagnosis in all groups (G1-G5), which were higher than those of prior studies reported [1, 13, 23, 25, 43]. Furthermore, the diagnoses made by pathologists using Morphogo system were consistent with the diagnoses of microscopic examination, as evidenced by the Kappa value of more than 0.85. It showed that Morphogo system is a useful tool for diagnosis of hematological diseases.

The study provided doctors with a potential way to apply AI to the morphology examination of BM smears. However, as the previous research reported, even for experienced pathologists, subtle differences between cells with similar morphological characteristics are difficult to identify [27]. This may explain why the sensitivity and PPV performance were not good in the identification of prolymphocyte, but the PPV of prolymphocyte and its similar cells was up to 1.0000. Furthermore, the image quality of BM nucleated cells depends on several factors, including the quality of BM smear preparation, the pathological condition and the imaging process [13]. This may be another cause of inaccurately identification of BM cells. The morphology of blasts in AL of G1 is more uniform, while they are relatively polymorphic and malformed in MDS of G2 [13]. Therefore, the blasts are easier to be identified and classified in AL, and difficult to be identified in MDS, which might be the cause of the higher misdiagnosis rate in some cases of Morphogo system compared with pathologists. Pathologists using Morphogo system made the diagnosis results of 16 cases were inconsistent with that made by conventional microscopes. Since the pathologists also refer to the results of other diagnostic tests when using microscope, such as flow cytometry and BM biopsy to diagnose disease. However, due to the massive number of BM samples every working day and the fact that BM cell differential count is so laborious and time-consuming, in some laboratories, while using Morphogo system they can review AI-based cell differential count results on the computer screen, which will dramatically improve the efficiency of laboratory work. We can imagine that Morphogo system can obtain and analyze more cells in a shorter period of time, and pathologists can review more cells to prevent some critical morphological changes from being missed, thus reducing the misdiagnosis rate and missed diagnosis rate. Meanwhile, it can help hospitals in rural area that lacks pathologists and save time for pathologists.

In this study, we used a single center method, in which all BM smears were prepared in the same laboratory and digitally processed. However, this study still had some limitations. The study was limited to 19 common diseases, and there were not enough number of cases for all of them. For example, there were only 4 cases of CML, 2 cases of CML-AP and lymphoma, so it was uncertain whether Morphogo system’s AI could have the same performance in all kinds of common hemato-pathological diseases and in even more rare conditions. In the future, more studies can be carried out with larger number of BM samples that consist of more types of hematological diseases from multiple laboratories to further validate the BM cell identification performance of Morphogo system.

The CNN-based Morphogo system can realize the recognition and classification of BM nucleated cells with high precision, thus providing a good platform for auxiliary pathologists to diagnose hematological diseases with high efficiency.

Ethics approval and consent to participate

The use of human samples was approved by the Ethics Committee of Hangzhou D.A. Medical Laboratory. Since BM aspirate smears in this retrospective study have been used in clinical examination, the informed consent of patients was exempted.

Consent for publication

Not applicable.

Availability of data and materials

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Competing Interests

All authors are employed by Hangzhou D.A. Medical Laboratory.

Funding

The authors did not receive support from any organization for the submitted work.

Authors’ contributions

Conceptualization: Qiufang Zhang; Methodology: Jingyuan Li and Yuan Peng; Formal analysis and investigation: Xiaofeng Zhang, Changhui Hua and Tingting Kong; Writing - original draft preparation: Qiufang Zhang; Writing - review and editing: Jingyan Wu; Resources: Yan Chen; Supervision: Yan Chen.

Huang F, Guang P, Li F, Liu X, Zhang W, Huang W (2020) AML, ALL, and CML classification and diagnosis based on bone marrow cell morphology combined with convolutional neural network: A STARD compliant diagnosis research. Medicine (Baltimore) 99 (45):e23154. https://doi.org/10.1097/MD.0000000000023154
Elsabah H, Soliman DS, Ibrahim F, Al-Sabbagh A, Yassin M, Moustafa A, Nashwan AM, Nawaz Z, ElOmri HM (2020) Plasma Cell Myeloma with an Aggressive Clinical Course and Anaplastic Morphology in a 22-Year-Old Patient: A Case Report and Review of Literature. Am J Case Rep 21:e920489. https://doi.org/10.12659/AJCR.920489
Dasariraju S, Huo M, McCalla S (2020) Detection and Classification of Immature Leukocytes for Diagnosis of Acute Myeloid Leukemia Using Random Forest Algorithm. Bioengineering (Basel) 7 (4). https://doi.org/10.3390/bioengineering7040120
Bain BJ, Bene MC (2019) Morphological and Immunophenotypic Clues to the WHO Categories of Acute Myeloid Leukaemia. Acta Haematol 141 (4):232-244. https://doi.org/10.1159/000496097
Wang SA, Hasserjian RP, Tam W, Tsai AG, Geyer JT, George TI, Foucar K, Rogers HJ, Hsi ED, Rea BA, Bagg A, Bueso-Ramos CE, Arber DA, Verstovsek S, Orazi A (2017) Bone marrow morphology is a strong discriminator between chronic eosinophilic leukemia, not otherwise specified and reactive idiopathic hypereosinophilic syndrome. Haematologica 102 (8):1352-1360. https://doi.org/10.3324/haematol.2017.165340
Fu X, Fu M, Li Q, Peng X, Lu J, Fang F, Chen M (2020) Morphogo: An Automatic Bone Marrow Cell Classification System on Digital Images Analyzed by Artificial Intelligence. Acta Cytol 64 (6):588-596. https://doi.org/10.1159/000509524
Gisslinger H, Jeryczynski G, Gisslinger B, Wölfler A, Burgstaller S, Buxhofer-Ausch V, Schalling M, Krauth MT, Schiefer AI, Kornauth C, Simonitsch-Klupp I, Beham-Schmid C, Müllauer L, Thiele J (2017) Clinical impact of bone marrow morphology for the diagnosis of essential thrombocythemia: comparison between the BCSH and the WHO criteria. Leukemia 31 (3):774-775. https://doi.org/10.1038/leu.2016.291
Chen P, Chen Xu R, Chen N, Zhang L, Zhang L, Zhu J, Pan B, Wang B, Guo W (2021) Detection of Metastatic Tumor Cells in the Bone Marrow Aspirate Smears by Artificial Intelligence (AI)-Based Morphogo System. Front Oncol 11:742395. https://doi.org/10.3389/fonc.2021.742395
Su J, Liu S, Song J (2017) A segmentation method based on HMRF for the aided diagnosis of acute myeloid leukemia. Comput Methods Programs Biomed 152:115-123. https://doi.org/10.1016/j.cmpb.2017.09.011
Chumachenko K, Iosifidis A, Gabbouj M (2022) Feedforward neural networks initialization based on discriminant learning. Neural networks : the official journal of the International Neural Network Society 146:220-229. https://doi.org/10.1016/j.neunet.2021.11.020
Shafique S, Tehsin S (2018) Computer-Aided Diagnosis of Acute Lymphoblastic Leukaemia. Comput Math Methods Med 2018:6125289. https://doi.org/10.1155/2018/6125289
Ehteshami Bejnordi B, Veta M, Johannes van Diest P, van Ginneken B, Karssemeijer N, Litjens G, van der Laak J, Hermsen M, Manson QF, Balkenhol M, Geessink O, Stathonikos N, van Dijk MC, Bult P, Beca F, Beck AH, Wang D, Khosla A, Gargeya R, Irshad H, Zhong A, Dou Q, Li Q, Chen H, Lin HJ, Heng PA, Haß C, Bruni E, Wong Q, Halici U, Öner M, Cetin-Atalay R, Berseth M, Khvatkov V, Vylegzhanin A, Kraus O, Shaban M, Rajpoot N, Awan R, Sirinukunwattana K, Qaiser T, Tsang YW, Tellez D, Annuscheit J, Hufnagl P, Valkonen M, Kartasalo K, Latonen L, Ruusuvuori P, Liimatainen K, Albarqouni S, Mungal B, George A, Demirci S, Navab N, Watanabe S, Seno S, Takenaka Y, Matsuda H, Ahmady Phoulady H, Kovalev V, Kalinovsky A, Liauchuk V, Bueno G, Fernandez-Carrobles MM, Serrano I, Deniz O, Racoceanu D, Venâncio R (2017) Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer. Jama 318 (22):2199-2210. https://doi.org/10.1001/jama.2017.14585
Wu YY, Huang TC, Ye RH, Fang WH, Lai SW, Chang PY, Liu WN, Kuo TY, Lee CH, Tsai WC, Lin C (2020) A Hematologist-Level Deep Learning Algorithm (BMSNet) for Assessing the Morphologies of Single Nuclear Balls in Bone Marrow Smears: Algorithm Development. JMIR Med Inform 8 (4):e15963. https://doi.org/10.2196/15963
Huang Z, Li Q, Lu J, Feng J, Hu J, Chen P (2021) Recent Advances in Medical Image Processing. Acta Cytol 65 (4):310-323. https://doi.org/10.1159/000510992
Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, Thrun S (2017) Dermatologist-level classification of skin cancer with deep neural networks. Nature 542 (7639):115-118. https://doi.org/10.1038/nature21056
Pattarone G, Acion L, Simian M, Mertelsmann R, Follo M, Iarussi E (2021) Learning deep features for dead and living breast cancer cell classification without staining. Sci Rep 11 (1):10304. https://doi.org/10.1038/s41598-021-89895-w
Tavakoli S, Ghaffari A, Kouzehkanan ZM, Hosseini R (2021) New segmentation and feature extraction algorithm for classification of white blood cells in peripheral smear images. Sci Rep 11 (1):19428. https://doi.org/10.1038/s41598-021-98599-0
Albarqouni S, Baur C, Achilles F, Belagiannis V, Demirci S, Navab N (2016) AggNet: Deep Learning From Crowds for Mitosis Detection in Breast Cancer Histology Images. IEEE transactions on medical imaging 35 (5):1313-1321. https://doi.org/10.1109/tmi.2016.2528120
Tang G, Fu X, Wang Z, Chen M (2021) A Machine Learning Tool Using Digital Microscopy (Morphogo) for the Identification of Abnormal Lymphocytes in the Bone Marrow. Acta Cytol 65 (4):354-357. https://doi.org/10.1159/000518382
Jin H, Fu X, Cao X, Sun M, Wang X, Zhong Y, Yang S, Qi C, Peng B, He X, He F, Jiang Y, Gao H, Li S, Huang Z, Li Q, Fang F, Zhang J (2020) Developing and Preliminary Validating an Automatic Cell Classification System for Bone Marrow Smears: a Pilot Study. J Med Syst 44 (10):184. https://doi.org/10.1007/s10916-020-01654-y
Zhang K, Zhang X (2021) Haemocyte variations in 35 species of grasshoppers and locusts. Sci Prog 104 (4):368504211053551. https://doi.org/10.1177/00368504211053551
Anwar SM, Majid M, Qayyum A, Awais M, Alnowami M, Khan MK (2018) Medical Image Analysis using Convolutional Neural Networks: A Review. J Med Syst 42 (11):226. https://doi.org/10.1007/s10916-018-1088-1
Kutlu H, Avci E, Ozyurt F (2020) White blood cells detection and classification based on regional convolutional neural networks. Med Hypotheses 135:109472. https://doi.org/10.1016/j.mehy.2019.109472
Bosse S, Maniry D, Muller KR, Wiegand T, Samek W (2018) Deep Neural Networks for No-Reference and Full-Reference Image Quality Assessment. IEEE Trans Image Process 27 (1):206-219. https://doi.org/10.1109/TIP.2017.2760518
Mori J, Kaji S, Kawai H, Kida S, Tsubokura M, Fukatsu M, Harada K, Noji H, Ikezoe T, Maeda T, Matsuda A (2020) Assessment of dysplasia in bone marrow smear with convolutional neural network. Sci Rep 10 (1):14734. https://doi.org/10.1038/s41598-020-71752-x
Goh KH, Wang L, Yeow AYK, Poh H, Li K, Yeow JJL, Tan GYH (2021) Artificial intelligence in sepsis early prediction and diagnosis using unstructured data in healthcare. Nat Commun 12 (1):711. https://doi.org/10.1038/s41467-021-20910-4
Matek C, Krappe S, Münzenmayer C, Haferlach T, Marr C (2021) Highly accurate differentiation of bone marrow cell morphologies using deep neural networks on a large image data set. Blood 138 (20):1917-1927. https://doi.org/10.1182/blood.2020010568
Seo MY, Hwang SJ, Nam KJ, Lee SH (2020) Significance of sleep stability using cardiopulmonary coupling in sleep disordered breathing. Laryngoscope 130 (8):2069-2075. https://doi.org/10.1002/lary.28379
Pereira KN, de Carvalho JAM, Paniz C, Moresco RN, da Silva JEP (2021) Diagnostic characteristics of immature platelet fraction for the assessment of immune thrombocytopenia. Thromb Res 202:125-127. https://doi.org/10.1016/j.thromres.2021.03.023
Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, Venugopalan S, Widner K, Madams T, Cuadros J, Kim R, Raman R, Nelson PC, Mega JL, Webster DR (2016) Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. Jama 316 (22):2402-2410. https://doi.org/10.1001/jama.2016.17216
Chen J, Kao YR, Sun D, Todorova TI, Reynolds D, Narayanagari SR, Montagna C, Will B, Verma A, Steidl U (2019) Myelodysplastic syndrome progression to acute myeloid leukemia at the stem cell level. Nature medicine 25 (1):103-110. https://doi.org/10.1038/s41591-018-0267-4
Suguna E, Farhana R, Kanimozhi E, Kumar PS, Kumaramanickavel G, Kumar CS (2018) Acute Myeloid Leukemia: Diagnosis and Management Based on Current Molecular Genetics Approach. Cardiovascular & hematological disorders drug targets 18 (3):199-207. https://doi.org/10.2174/1871529x18666180515130136
Ohanian M, Kantarjian HM, Shoukier M, Dellasala S, Musaelyan A, Nogueras Gonzalez GM, Jabbour E, Abruzzo L, Verstovsek S, Borthakur G, Ravandi F, Garcia-Manero G, Tamamyan G, Champlin R, Pierce S, Ferrajoli A, Kadia T, Cortes JE (2020) The clinical impact of time to response in de novo accelerated-phase chronic myeloid leukemia. Am J Hematol. https://doi.org/10.1002/ajh.25907
Wickramasinghe SN (2006) Diagnosis of megaloblastic anaemias. Blood reviews 20 (6):299-318. https://doi.org/10.1016/j.blre.2006.02.002
Palomo L, Acha P, Solé F (2021) Genetic Aspects of Myelodysplastic/Myeloproliferative Neoplasms. Cancers (Basel) 13 (9). https://doi.org/10.3390/cancers13092120
Saygin C, Carraway HE (2021) Current and emerging strategies for management of myelodysplastic syndromes. Blood reviews 48:100791. https://doi.org/10.1016/j.blre.2020.100791
Deininger MW (2015) Diagnosing and managing advanced chronic myeloid leukemia. American Society of Clinical Oncology educational book American Society of Clinical Oncology Annual Meeting:e381-388. https://doi.org/10.14694/EdBook_AM.2015.35.e381
Han P, Hou Y, Zhao Y, Liu Y, Yu T, Sun Y, Wang H, Xu P, Li G, Sun T, Hu X, Liu X, Li L, Peng J, Zhou H, Hou M (2021) Low-dose decitabine modulates T-cell homeostasis and restores immune tolerance in immune thrombocytopenia. Blood 138 (8):674-688. https://doi.org/10.1182/blood.2020008477
Hallek M (2019) Chronic lymphocytic leukemia: 2020 update on diagnosis, risk stratification and treatment. Am J Hematol 94 (11):1266-1287. https://doi.org/10.1002/ajh.25595
Auerbach M, Adamson JW (2016) How we diagnose and treat iron deficiency anemia. Am J Hematol 91 (1):31-38. https://doi.org/10.1002/ajh.24201
Sun L, Babushok DV (2020) Secondary myelodysplastic syndrome and leukemia in acquired aplastic anemia and paroxysmal nocturnal hemoglobinuria. Blood 136 (1):36-49. https://doi.org/10.1182/blood.2019000940
Brown AL, Hahn CN, Scott HS (2020) Secondary leukemia in patients with germline transcription factor mutations (RUNX1, GATA2, CEBPA). Blood 136 (1):24-35. https://doi.org/10.1182/blood.2019000937
Su J, Han J, Song J (2021) A benchmark bone marrow aspirate smear dataset and a multi-scale cell detection model for the diagnosis of hematological disorders. Comput Med Imaging Graph 90:101912. https://doi.org/10.1016/j.compmedimag.2021.101912

Competing interest reported. All authors are employed by Hangzhou D.A. Medical Laboratory.

Application of Morphogo based on convolutional neural network for morphological identification of bone marrow nucleated cells

Status:

Version 1

Abstract

Introduction:

Methods

Results

Conclusion

Figures

1. Introduction

2. Methods

2.1 Sources and Classification of Sample

2.2 CNN construction

2.3 Data digitization

2.4 Morphogo system workflow

2.5 Morphogo system evaluation

2.6 Statistical analysis

3. Results

3.1 The Morphogo system is capable of accurately identifying BM nucleated cells

3.2 The Morphogo system and the pathologist are extremely consistent in identifying BM nucleated cells

3.3 The application of Morphogo system can effectively diagnose hematological system diseases.

4. Discussion

5. Conclusion

Declarations

References

Additional Declarations

Status:

Version 1