Leveraging Artificial Intelligence for Improved Hematologic Cancer

Care: Early Diagnosis and Complications’ Prediction

Yousra El Alaoui

, Regina Padmanabhan

, Adel Elomri

, Halima El Omri

and Abdelfatteh El Omri

College of Science and Engineering, Hamad Bin Khalifa University, Doha, Qatar

National Center for Cancer Care and Research, Hamad Medical Corporation, Doha, Qatar

Surgical Research Section, Department of Surgery, Hamad Medical Corporation, Doha, Qatar

Keywords: Cancer Research, Machine Learning, Hematology Management.

Abstract: Today, medical artificial intelligence (AI) applications are being extensively utilized to enhance the outcomes

of clinical diagnosis and overall patient care. This data-driven approach can be trained to account for

individuals’ unique characteristics, medical history, ethnicity, and even genetic make-up to obtain accurately

tailored treatment recommendations. Given the power of medical AI, the severe nature of hematological

malignancies and the related constraints in terms of both time and cost, in this paper, we are investigating the

importance of AI applications in hematology management, with an illustration of AI’s role in reducing pre-

and post-diagnosis challenges. Insights discussed here are derived based on our experiments on clinical

datasets from National Center for Cancer Care & Research (NCCCR), Qatar. Specifically, we developed AI

models for blood cancer diagnosis as well as prediction of therapy-induced clinical complications in patients

with hematological cancers to facilitate better hospital management and improved cancer care.

1 INTRODUCTION

Artificial intelligence (AI) is the science of using

computer systems and algorithms that simulate the

human brain to perform tasks and solve complex

problems.

Thanks to the recent explosive improvement and

progress in computing power and easy access to large

repositories of data, the realm of AI currently plays a

big role in several technologies, including clinical

medicine and biomedical research (Radakovich et al.,

2020).

Driven by the abundance of medical data and the

remarkable AI results in various fields and machine

learning (ML), there exist several promising tools that

can help clinicians solve critical problems related to

oncology and hematology (El Alaoui et al., 2022).

Examples of successful AI applications in

cancer research range from digital medical image

analysis and pattern recognition to cancer

https://orcid.org/0000-0002-7389-2789

https://orcid.org/0000-0001-9448-6950

https://orcid.org/0000-0003-1605-9800

https://orcid.org/0000-0002-8299-3733

https://orcid.org/0000-0003-4112-7924

classification and diagnosis using ML and deep

learning (DL) algorithms (Walter et al., 2021).

2 AI IN HEMATOLOGY

MANAGEMENT

With the improved access to ML and DL tools and

medical data augmentation and expansion techniques,

the scope of AI applications in hematology has

increased to include all stages of patient management

from diagnosis to treatment and prognosis (Shadman

et al., 2023). Despite recent technical advancements

and variety of models developed to support

hematological data, AI research in hematological

cancer management was found to receive less

attention compared to oncology (El Alaoui et al.,

2022). Furthermore, the current state-of-the-art

emphasized on some hematological malignancies

El Alaoui, Y., Padmanabhan, R., Elomri, A., El Omri, H. and El Omri, A.

Leveraging Artiﬁcial Intelligence for Improved Hematologic Cancer Care: Early Diagnosis and Complications’ Prediction.

DOI: 10.5220/0012360900003657

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2024) - Volume 2, pages 87-90

ISBN: 978-989-758-688-0; ISSN: 2184-4305

more than others, namely Acute Myeloid Leukemia

(AML) and Acute Lymphoblastic Leukemia (ALL)

compared to Chronic Myeloid Leukemia (CML) and

Chronic Lymphoblastic Leukemia (CLL), wherein

further research is pivotal (El Alaoui et al., 2022).

Table 1 portrays the results of a literature review

encompassing 131 papers on the topic of AI

applications in hematology categorized by

malignancy type (El Alaoui et al., 2022).

Table 1: Categorization of 131 papers by malignancy type

(N=131).

Malignancy type Values, n (%)

Acute myeloid leukemia

(AML)

42 (32.1)

Acute lymphoblastic

leukemia (ALL)

40 (30.5)

Chronic lymphocytic

leukemia (CLL)

13 (9.9)

Chronic myeloid

leukemia (CML)

2 (1.5)

Lymphoma

Others

17 (13.0)

Contrary to solid cancers, the diagnosis of liquid

cancers is considered more challenging and time-

consuming, given the complex symptomatology of

the disease and the absence of symptoms in suspected

patients during early stages, which restricts the

clinicians’ ability to predict the occurrence of such

disease beforehand. While traditional detection

methods rely mainly on blood cell image

classification, AI-based models were introduced to

enhance identification accuracy and provide a pilot

view on the potential spread of the disease within the

patient’s body to help improve the chances of patient

recovery and increase survival rates (El Alaoui et al.,

2022).

Currently, the initial causes of blood cancers like

leukemia are unknown, and no screening tests have

been proven efficient enough to diagnose blood

cancer in its early stages, unless some warning signs

develop overtime (El Alaoui et al., 2022). In addition

to the ambiguity of blood cancer signs, many patients

do not exhibit any symptoms at the time of diagnosis,

leading to delays in disease detection and treatment

initiation. For instance, 80% of Chronic

Lymphoblastic Leukemia (CLL) patients are

asymptomatic at the time of diagnosis (Shadman et

al., 2023). Such pre-diagnosis challenges faced by

clinicians highlight the importance of AI-based

models in tackling delayed diagnosis and the resulting

implications. Moreover, applications of AI in the

treatment stage and beyond, have yielded interesting

outcomes in terms of achieving a timely and efficient

therapy, and help predicting post-diagnosis

complications (multiple infections, myelosuppression

etc.) for a better patient management flow. Figure 1

summarizes the different pre- and post-diagnosis

challenges and how AI can be used for combatting

these challenges.

Figure 1: Advancing Healthcare Safety for Hematological

Cancer Patients through AI.

As patients with hematological malignancies

have an impaired immune function (immune

suppression), they often report to hospitals with

repeated multiple infections. Sometimes these

patients develop high fever and reduced neutrophil

count (febrile neutropenia-FNE). Reportedly, patients

with FNE are very likely to develop fatal sepsis and

thus mortality rate is high in such patients. As shown

in Figure 1, AI-driven outcome prediction models can

be used to identify such patients and give them

required care from the beginning of hospitalization

itself. Another post-diagnosis challenge is subclass

determination; as treatment options vary considerably

for each subclass hematological malignancy, highly

specific molecular and cytological tests are required

to identify exact subclass. All these challenges

complicate hematological cancer care.

To address both these challenges, we are presenting

two case studies, (a) ML-based models for early

diagnosis of leukemia and (b) and ML-models to

predict potential post-diagnosis complications in

patients with hematological malignancies,

respectively.

3 DISCUSSION

3.1 Case Study (a): Diagnostic Systems

for Early Leukemia Detection

With the aim of enhancing Acute Lymphoblastic

Leukemia (ALL)’s detection accuracy and

HEALTHINF 2024 - 17th International Conference on Health Informatics

overcoming the challenges associated with

conventional manual microscopic identification

techniques, we developed an AI-based diagnostic

system for ALL detection using Random Forest (RF),

XGBoost (XGb) and Decision Tree (DT) algorithms

(El Alaoui et al., 2023). The three models were

trained using 86 ALL and 86 control patients.

Moreover, a Grid Search hyperparameter tuning

technique was applied on each of the three

classification models, while a Forward Feature

Selection approach was performed to select the 10

most ALL-discriminatory complete blood count

(CBC) features out of the initial 20, which included:

Absolute Neutrophil Count (ANC), Hematocrit

(HCT), Red Blood Cells (RBCs), MCV, MCH,

Neutrophils %, Basophils %, Lymphocyte count,

MPV and Platelets. Training the models using a 5-

fold cross-validation technique resulted in high

accuracies corresponding to 91.4% for DT and an

identical 88.6% for both RF and XGb, respectively

(El Alaoui et al., 2023).

We also developed a ML-based diagnosis and

screening model for Chronic Lymphoblastic

Leukemia (CLL) using 3 ML techniques including

Linear Regression (LR), Linear Discriminant

Analysis (LDA) and XGb, selected out of 8 candidate

models following 5-fold cross-validation. The three

models were trained using a total of 682 CBC records,

where 88 were confirmed CLL patients and 594 were

control. Dataset imbalance and disproportionality

were overcome by means of (SMOTE-Synthetic

Minority Over-Sampling Technique), (Padmanabhan

et al., 2023). Moreover, the common high-ranked

CBC parameters were extracted using chi-square,

mutual information, extra tree and XGboost

classifiers. The final set of features included WBC,

Lymphocyte count, Neutrophil %, Lymphocyte %,

Monocyte %, ANC, Platelets, Basophil %, Monocyte

count, Basophil count and Eosinophil count, by order.

The final performances of the selected models using

the 11 chosen CBC parameters resulted in a 97.05%,

97.63% and 98.62% accuracy corresponding to QDA,

LR and XGb, respectively (Padmanabhan et al.,

2023).

Based on these novel results from the two

aforementioned studies, it is fair to mention that pre-

diagnosis delays can be effectively tackled through

the integration of AI models in diagnostic systems

and devices to foster early hematological disease

detection.

3.2 Case Study (b): Prediction of

Clinical Complications in Cancer

Patients

With the aim of expediating the medical diagnosis

process and enhancing its reliability for better health

outcomes and increased chances of patient survival,

we developed a novel AI-based approach to better

manage patients with hematological cancer, namely

the ones affected by therapy-induced

myelosuppression, multiple infections, and febrile

neutropenia (FN) (Padmanabhan et al., 2022).

In addition to the increasing sepsis risk and

mortality in hematologic cancer patients with FN

associated with treatment-induced myelosuppression,

a high prevalence of multidrug-resistant organisms is

also captured in such patients, which limits the

number of treatment options that the patient is

allowed amidst a similar set of health complications.

Therefore, the early identification of such organisms

within the cancer patient body can help prepare the

latter to receive a better treatment, enable good

hospital management and prevent the spread of such

organisms to the weaker category of patients

(Padmanabhan et al., 2022).

The current application serves as a predictive

model for multidrug-resistant organisms, sepsis, and

mortality risk in hematological cancer patients with

FN. The application consists of medical data

extraction using 1166 febrile neutropenia episodes

reported in 513 patients, trained using the XGboost

algorithm, a model selected out of 6 candidate models

using a 10-fold cross-validation. Furthermore, to

address the data disproportionality problem, data

augmentation techniques and model-scoring-based

hyperparameter tuning were used, and a set of

features were added to the model to enhance the

predictability of the previously mentioned clinical

complications. The performance for sepsis, multi-

drug resistant organism and mortality predictions

resulted in a 0.85, 0.91 and 0.88 AUC (area under the

curve), respectively, highlighting AI’s potential in

treatment clinical decision-making. Figure 2

represents the adopted methodology followed in

building the aforementioned predicted models.

4 CONCLUSIONS

The computational power of AI constitutes a very

strong leap in the field of hematology management,

such that it enables a deep level of understanding of

the unseen relations between clinical patient

Leveraging Artiﬁcial Intelligence for Improved Hematologic Cancer Care: Early Diagnosis and Complications’ Prediction

Figure 2: Schematic representation of the adopted

methodology for building predictive models.

parameters and patient outcomes. Indeed, this

powerful basis sets the ground for further

advancements in the realm of healthcare management

and upscaling of cancer care. Nevertheless, some

ethical considerations that are sought to guide clinical

decision making amidst medical AI applications are

yet to be discussed.

ACKNOWLEDGEMENTS

This article was made possible by National Priorities

Research Program-Standard (NPRP-S) Twelfth

(12th) Cycle grant# NPRP12S-0219-190108, from

the Qatar National Research Fund (a member of the

Qatar Foundation). The findings herein reflect the

work, and are solely the responsibility, of the

author[s].

REFERENCES

Radakovich, N., Nagy, M., & Nazha, A. (2020a).

Artificial Intelligence in hematology: Current

challenges and opportunities. Current Hematologic

Malignancy Reports, 15(3), 203–210. https://doi.org/

10.1007/s11899-020-00575-4

El Alaoui, Y., Elomri, A., Qaraqe, M., Padmanabhan, R.,

Yasin Taha, R., El Omri, H., EL Omri, A., &

Aboumarzouk, O. (2022). A Review of Artificial

Intelligence Applications in Hematology Management:

Current Practices and Future Prospects.

https://doi.org/10.2196/preprints.36490

Walter, W., Haferlach, C., Nadarajah, N., Schmidts,

I., Kühn, C., Kern, W., & Haferlach, T. (2021).

How artificial intelligence might disrupt diagnostics

in hematology in the near future. Oncogene,

40(25), 4271–4280. https://doi.org/10.1038/s41388-

021-01861-y

Shadman, M. (2023). Diagnosis and treatment of chronic

lymphocytic leukemia. JAMA, 329(11), 918.

https://doi.org/10.1001/jama.2023.1946

El Alaoui, Y., Padmanabhan, R., Elomri, A., Qaraqe,

M. K., El Omri, H., & Yasin Taha, R. (2023).

An artificial intelligence-based diagnostic system

for acute lymphoblastic leukemia detection.

Studies in Health Technology and Informatics.

https://doi.org/10.3233/shti230479

Padmanabhan, R., El Alaoui, Y., Elomri, A., Qaraqe, M. K.,

El Omri, H., & Yasin Taha, R. (2023). Machine

learning for diagnosis and screening of chronic

lymphocytic leukemia using routine complete blood

count (CBC) results. Studies in Health Technology and

Informatics. https://doi.org/10.3233/shti230483

Padmanabhan, R., Elomri, A., Taha, R. Y., El Omri,

H., Elsabah, H., & El Omri, A. (2022). Prediction

of multiple clinical complications in cancer patients

to ensure hospital preparedness and Improved

Cancer Care. International Journal of Environmental

Research and Public Health, 20(1), 526.

https://doi.org/10.3390/ijerph20010526

HEALTHINF 2024 - 17th International Conference on Health Informatics