| Journal of Visual Artificial Intelligence
Received: 08 August 2026; Revised: 16 September 2026; Accepted: 21 September 2026; Published Online: 22 September 2026.
J. Vis. Artif. Intell., 2026, 1(2), 26107 | Volume 1 Issue 2 (September 2026) | DOI: https://doi.org/10.64189/vai.26107
© The Author(s) 2026
This article is licensed under Creative Commons Attribution NonCommercial 4.0 International (CC-BY-NC 4.0)
Q-MedDict: A Classical Quantum-Inspired
Dictionary for Lung Tissue Images and Laboratory
Metadata
Sazia Parvin
*
and Md Sarwar Kamal
*
Melbourne Polytechnic Melbourne, VIC, 3072, Australia
*Email: SaziaParvin@melbournepolytechnic.edu.au (Sazia Parvin); mdsarwarkamal@melbournepolytechnic.edu.au (Md Sarwar
Kamal)
Abstract
In the age of the information superhighway, artificial intelligence is assisting in health care and other medical
domains. Considering the usefulness of AI, it would be beneficial to have automated lung cancer identification
based on various cancer data. These data could be images, tabular data, or handwritten notes. Lung cancer data
may include images and laboratory findings, and may have different structures and scopes. In this research, we
present a classical quantum-based dictionary representation to handle various combinations of data. The
quantum dictionary facilitates the superposition principle of quantum computing. It does not run on quantum
hardware but is instead considered as a logical representation of a quantum dictionary using superposition. We
have used dictionary framework with quantum gates such as H, S, T, CX, and CCX with matrix formation.
Moreover, we utilize discrete cosine transform (DCT) and Karhunen–Loève transform (KLT) image sources. We
found that our proposed approach shows better performance at 25% coefficient value improvements. Moreover,
quantum dictionary-based key and value pairs ensure Hadamard-centric image formations up to 30.85 dB and
24.34 dB peak signal-to-noise ratio (PSNR). Similarly, proposed approach performs over other image sources
like DCT up to 32.77 dB and 26.38 dB.
Keywords: Quantum-inspired dictionary; Classical simulation; Digital pathology; Lung H&E, Laboratory metadata;
NHANES; Hadamard transform; Multimodal representation.
1. Introduction
Explainable artificial intelligence (XAI) and quantum computing (QC) offer beneficial advantages in cancer data
analytics.
[1]
Data of disease such as cancer data can include various data types. This data can be XRAY, images,
laboratory sample findings, practisioner notes and text.
[2]
A pathologist can look at a slide of stained tissue. A
doctor can also look at your blood and urine tests. Such combinations of data are known as multimodal data.
These data don’t seem to be alike. An image can be made of millions of pixels. A laboratory table can only have
a few values. Thus, a multimodal system needs to be able to make the source of each value visible.
[3]
Multimodal
data analytics would provide insights about insights about cancer disease.
[4]
One of the key initialization is
process data digitally. In such, digital pathology is growing quickly for medical data analytics. In this case, XAI
and QC would be effective to handle large amount of data.
[5]
Recent lung pathology models can read many tissue patterns from images. Other work moves toward
multimodal pathology. MuPD places histology, RNA, and text in a shared space.
[6]
H
2
O integrates Hematoxylin
and Eosin (H& E) stained tissue images with spatial multi-omics data.
[7]
SpatialFusion combines tissue,
expression, and pathway information.
[8]
These models use learned representations and large training sets.
There is also growing interest in quantum computing and its applications in medical data analysis.
[9]
Quantum
image representation methods, such as Flexible Representation of Quantum Images (FRQI) and Novel Enhanced
Quantum Representation (NEQR), encode image information into quantum states.
[10,11]
The Quantum Pixel
Representation (QPIXL) framework was used to support quantum image analysis and manage circuit data.
[12]
From the study, we see that our proposed approach and these approaches are not same. We find a few research
gaps in the existing literature. These gaps are related to processing multimodal data using a classical model, the
black-box nature of the models, and the lack of interpretability.
[13]
Black-box models are those that perform
their operations through a large number of iterations, making it difficult to understand and track how the
operations are concluded.
[14,15]
In this case, we only get the output, not an understanding of how it was obtained.
However, in medical research, not all experts have an AI background. Therefore, we need interpretable AI
models.
[16]
Similarly, there are some other gaps related to the complexities of quantum computing interactions.
To avoid quantum complexities, we introduce dictionary-based quantum computing for processing larger
datasets as well as improving interpretability. The public sources are important. Cancer Moonshot Biobank–
Lung Cancer Collection (CMB-LCA) is a Cancer Moonshot Biobank lung cancer collection hosted by The Cancer
Imaging Archive (TCIA) and represented in Imaging Data Commons (IDC).
[17,18]
The National Health and
Nutrition Examination Survey (NHANES) is a separate public survey. Its Complete Blood Count (CBC) file
includes WBC, haemoglobin, and platelet measures.
[19]
Its biochemistry file includes albumin, BUN, creatinine,
and glucose.
[20]
Its urine file includes urine albumin, urine creatinine, and albumin-tocreatinine ratio.
[21]
In this
research we have the following contributions:
1) A classical quantum-centric dictionary with explicit bitstring keys and amplitudes.
2) A direct numerical demonstration of all five stated dictionary rules: H, S, T, CX, and CCX.
3) A lung H&E block experiment with Hadamard, DCT, and PCA/KLT references. To prove the hypothesis, we
consider a research question, which is: Can a classical quantum-based dictionary form design lung cancer image
and pathological metadata in a reliable way, while considering sparse representation to be calculated
separately?
2. Related work
Cancer image may contain small cell details and large tissue regions (Fig. 1). A workflow is to cut the slide into
tiles. Each tile is translated into features. CLAM used attention-based multiple instance learning for weakly
supervised image analysis.
[22]
Current foundation models are trained on larger collections of pathology.
PulmoFoundation focuses on lung pathology.
[23]
Foundation models for pathology also suffer from issues like
stain variation, site variation, data size, and clinical Validation.
[24]
Here these problems are critical because two
demonstration images cannot show the entire range of lung pathology. MuPD integrates histology, RNA and
text.
[25]
H
2
O and SpatialFusion bridge tissue with molecular data.
[26,27]
OmiCLIP connects histology and spatial
transcriptomics.
[7]
These methods utilize funcitons centric data and find hidden imag spaces. Q-MedDict is
dictionary-based approach. In this method, combines a specific image pixel with respect to a desired location
which is a key. Each key can show a modality and a feature position. So, our proposed method is a lightweight
model that performs well on multimodal data within quantum scopes. In particular, the quantum features with
superposition make it a strong candidate for handling multimodal data in cancer data analysis. FRQI stores
image colour information in probability amplitudes and image position in basis states.
[28]
NEQR stores grayscale
values in computational basis states.
[29]
QPIXL provides a common circuit framework for several quantum pixel
representations.
[30]
These approaches are regular quantum-state representations or circuit foundations. One the
other hand, our proposed Q-MedDict is different as this is dictionary-based data mapping. It is a classical
dictionary-based data structure using quantum notation. In this approach, we do not need a quantum processor.
The data orientations are dictionary keys and values pair. The quantum gate functions rules are general
deterministic update functions. According to the a newly proposed method, image color values data are Image
color information is stored in probability amplitudes.
[31]
Similarly, a grayscale values are stored in computational
basis states.
[32]
Another research find out a generic circuit structure for many quantum pixel representations is
provided by QPIXL.
[12]
Our proposed method QMedDict is not the same. It is a traditional data format that draws
inspiration from quantum notation. The bit strings are standard dictionary keys.
Fig. 1: Lung adenocarcinoma image.
2.1 Positioning against baselines
Table 1 presents the descriptions of existing approaches and our proposed approach. DCT works on image
transform. Other methods such as PCA/KLT is a data-centric linear transform. On the other hand, FRQI and
NEQR are quantum image representations. Our proposed, Q-MedDict offer key and value combinaitons to trace
each data with their locations.
3. Data and study design
In this research, we used publicly available data. One of the data is CMB-LCA is a public lung cancer collection
from the Cancer Moonshot Biobank. All these data.
[17]
IDC lists slide microscopy as one of its modalities.
[18]
3.1 NHANES blood and urine data
We have collected the laboratory data from public NHANES documentation for the August 2021 to August 2023
release. Table 2 presents the summaries of the metadata we used in this research from laboratory data.
These are:
• CBC_L: WBC, haemoglobin, platelet count and other blood-cell measures.
• BIOPRO_L: albumin, BUN, serum creatinine, glucose and other chemistry measures.
• ALB_CR_L: urine albumin, urine creatinine, and albumin/ creatinine ratio.
4. Q-Meddict method
Our proposed method is an integrated approach of multiple processes. The proposed framework shown in Fig.
2. In this figure we see that, first step is about data collections and then normalizations and so on until the last
state which is re-building the image. A state is stored as
(1)
The key
is a bit sequence. The amplitude
present the amount connected to that key. For an image patch, a
key can show a pixel location. For a table, a key can mark a feature slot. In Table 3, we present quantum
dictionary based gates and rules which account for quantum data processing.
The amplitudes are presented by
,
which gives
(3)
Table 1: Descriptions of various methods along with our proposed method
Approach
Data types
Methods
Advantages
Variations
DCT
[33]
Graphical image data
Cosine similarity
Classical computing
for image data
Modality based
PCA/KLT
Vector/image
Segments
Geometrical
structure
Linear formation
Gate based analysis
FRQI
[28]
Image and graphical
data
Quantum computing
Image on quantum
state
Needed a quantum computing
design
NEQR
[29]
Image data
Gray scale image
encoding
Grayscale basic design
Quantum image design
QPIXL
[12]
Image data
Quantum pixel design
Quantum image
circuit approach
Circuit based
Q-MedDict
Graphical image data
and tabular lab data
Dictionary based
design
Keys and values
A small representation
analysis
Fig. 2: Q-MedDict framework.
Table 2: NHANES release metadata used in the laboratory branch. these are codebook ranges and counts, not patient
measurements
Type
Variable
Public value range
Valid records
Availability
Blood
WBC (1000 cells/µL)
2.1–20.8
7,593
87.01%
Blood
Haemoglobin (g/dL)
6.3–18.6
7,593
87.01%
Blood
Platelets (1000 cells/µL)
46–787
7,593
87.01%
Blood
Serum albumin (g/dL)
2.2–5.5
6,366
88.43%
Blood
BUN (mg/dL)
4–74
6,326
87.87%
Blood
Serum creatinine (mg/dL)
0.33–15.17
6,326
87.87%
Blood
Serum glucose (mg/dL)
40–898
6,357
88.30%
Urine
Urine albumin (µg/mL)
0.01–7942.47
8,153
96.00%
Urine
Urine creatinine (mg/dL)
4–873
8,154
96.01%
Urine
Albumin/creatinine ratio (mg/g)
0–14708.28
8,153
96.00%
4.1 Image branch
Each lung H&E image is presented to grayscale for the present value. The image is represented as a 256× 256
array and split into 16×16 blocks. A block contains 256 values.
For compact notation, the orthonormal Hadamard gate is
, (4)
where is one image location and is a normalised Hadamard table. The largest are kept:
otherwise.
(5)
The block is supported by
(6)
We test = 128, 64, and 32. These values keep 50%, 25%, and 12.5% of block coefficients.
Our proposed key and dictionary quantum approach support pairwise gate update to key and value
combinations. The matrix structure is used because it is easy to execute and get output. We verified dictionary
values and the reference matrix output in Table 4. We also report the fraction of squared transform coefficients
kept:
Table 3: Dictionary rules used for Q-MedDict
Rule
Simple role in the dictionary
H
Mix a pair of amplitudes as
and
.
S
Multiply amplitudes whose target bit is 1 by .
T
Multiply amplitudes whose target bit is 1 by
.
CX
Flip a target key bit when one control key bit is 1.
CCX
Flip a target key bit when two control key bits are 1.
Table 4: Numerical verification of the dictionary rules.
Rule
Max. absolute difference
H
1.11 × 10
−16
S
0
T
0
CX
0
CCX
0
Here we see that due to Hadamad gate H is orthonormal, the full energy before and after the transform is the
same. Therefore, RE is not supporting that the transform “preserves energy.” It measures how reserved the
coefficients are after top-K selection.
4.2 Laboratory metadata branch
We define the experiments as follows:
(8)
and a release completeness value
The reserved descriptor used
before min and max scaling and amplitude normalization.
Our interpretation as follows: where
and
are descriptors. We also do not recommend multiplying a future
patient measurement by qi. We define a record is
. We define a record is
(10)
where
is a scaled value,
is a missing value, and
is a metadata.
4.4 Gate-like dictionary rules
Q-MedDict uses classical update rules inspired from quantum computing rules.
5. Experimental setup
In these experiments we used two public lung adenocarcinoma images. Each image was divided into 256 groups
of 16 × 16 pixels. Hadamard values were managed by true size. These are 128, 64, or 32. For a base line outcome,
we used same arrays and the same top-K rule with a two-dimensional orthonormal DCT image. A PCA/KLT
values was also used. For PCA/KLT, one image was used to the 512 blocks from the two demonstration images.
The storage cost is not measured.
Table 5: Lung H&E image results for the Hadamard-gate
Image
Keep (%)
PSNR (dB)
SSIM
Energy (%)
H&E-1
50.0
37.60
0.983
99.96
H&E-1
25.0
30.85
0.919
99.80
H&E-1
12.5
27.47
0.814
99.56
H&E-2
50.0
31.24
0.978
99.87
H&E-2
25.0
24.34
0.910
99.35
H&E-2
12.5
20.83
0.809
98.58
Table 6: Image baselines at 64 of 256 coefficients per block (25%)
Image
Method
PSNR (dB)
SSIM
Image
Method
H&E-1
Hadamard
30.85
0.919
H&E-1
Hadamard
H&E-1
DCT
32.77
0.947
H&E-1
DCT
H&E-1
PCA/KLT
28.89
0.864
H&E-1
PCA/KLT
H&E-2
Hadamard
24.34
0.910
H&E-2
Hadamard
H&E-2
DCT
26.38
0.942
H&E-2
DCT
H&E-2
PCA/KLT
25.04
0.910
H&E-2
PCA/KLT
5.1 Laboratory metadata experiment
The NHANES instructions were utilized to record the minimum, maximum, valid count, and total count. Seven
blood descriptors and three urine descriptors were also used. We also used “release-metadata framework.”
Moreover, We also added a numerical sensitivity check. It checks the initial
descriptor with: (i) weighting,
(ii) no log transform, and (iii) identity top-K keeping.
5.2 Scope rule
We define the image and other metadata for our experimental analysis. We consider these data to investigate
our hypothesis weather quantum dictionary-based model can have better utilization of superposition.
6. Results
6.1 Lung image encoding results
Table 5 presents the original Hadamard-gate centric image outcomes. Most values are representing the image
data analysis with various combinations. Table 6 presents further outcomes of 25% baseline comparative
analysis. The DCT is perform better outcomes in comparison with the Hadamard for PSNR and SSIM images.
Similarly, the PCA/KLT is shows in between outcomes. QMedDict offers better scopes for dictionary
representation and rule tracing. Fig. 3 shows grayscale image and its 25% Hadamard framework. The largest
framework remains usable. Fig. 4 shows that PSNR falls when fewer coefficients are kept. The dictionary values
are irregular. Fig. 5 shows the 20 largest values from one tissue patch.
Fig. 3: Lung image which is rebuilt after keeping 64 of 256 Hadamard coefficients in each 16 × 16 block.
Fig. 4: PSNR against the share of kept Hadamard coefficients.
Fig. 5: Top 20 normalised dictionary amplitudes from one lung H&E patch.
6.2 Blood release-metadata results
Fig. 6 shows the comparative outcomes for completeness values. Blood variables are around 87 to 88%
complete in the selected files. Urine variables are around 96%. Table 7 presents descriptor reconstruction
numbers which are shows the exact coefficient counts. It also shows the error as RMSE because min and max
scaling values. These numbers describe an eight-element blood metadata vector and a four element urine
metadata vector. Fig. 7 shows reconstruction of the seven-value release descriptor.
Fig. 6: Share of available records reported in the selected NHANES codebooks.
Table 7: Original release-metadata dictionary reduction results
State
Coeff. kept
Keep (%)
RMSE
Cosine
Energy (%)
Blood
4/8
50
0.068
0.993
98.64
Blood
2/8
25
0.239
0.911
82.53
Urine
2/4
50
0.089
0.993
98.53
Urine
1/4
25
0.454
0.810
49.26
Fig. 7: Blood release-metadata descriptor before and after 50% Hadamard coefficient keeping.
6.3 Urine release-metadata results
The urine data analysis shows 8,154 creatinine results, and 8,153 creatinine ratio results. Fig. 8 shows a
metadata-vector framework.
Fig. 8: Urine release-metadata descriptor.
6.4 Cross-branch coefficient concentration
Fig. 9 shows comparative outcomes of original and re-build lab data with respect to our proposed method. From
the figure we see that, using quantum dictionary-based analysis produces better formations.
Fig. 9: top-K squared-coefficient.
6.5 Answer to the research question
Our analysis shows that the answer lies at the data-structure level. A classical key–value schema can store both
the retained image coefficients and the laboratory release values, and the dictionary combinations built on it
ran successfully in all experiments.
6.6 Sensitivity to the number of kept states
Fig. 10 shows the original SSIM change for the two lung images. At 50% coefficient keeping, both images were
above 0.97 SSIM. At 25%, both were above 0.90. At 12.5%, both were close to 0.81. Fig. 11 shows the public
range widths after a log display. It is useful for seeing scale differences. It does not describe within-patient
variation.
6.7 Example of one future multimodal dictionary record
A new analysis could use separate key value combinations. Image keys may start with 00, blood keys with 01,
and urine keys with 10. For example,
(11)
Fig. 10: SSIM against the share of kept Hadamard image coefficients.
Fig. 11: Selected blood and urine variables.
Table 8: Numerical sensitivity of the 50% release-metadata descriptors
State
Variant
RLE
Cosine
Blood
Initial log range ×q
0.117
0.993
Blood
No availability weight
0.112
0.994
Blood
No log transform
0.087
0.996
Blood
Identity top-K
0.207
0.978
Urine
Initial log range ×q
0.121
0.993
Urine
No availability weight
0.121
0.993
Urine
No log transform
0.308
0.951
Urine
Identity top-K
0.000
1.000
The prefix identifies the modality. The remaining bits identify a feature. A separate mask bit or mask field can
record whether a value was observed.
6.8 Quantitative descriptor sensitivity
Table 8 shows the qualitative values for validations. The outcomes indicate the metadata should not be
overestimated. For urine, identity top-K keeping reconstructs the small padded descriptor. This indicates that
the better retention number is a property of the positive vector and its padding.
7. Discussion
This analysis shows that Hadamard variations framework have useful outcomes for the two demonstration
images with small values to high similarity values. On the other hand, DCT baseline provides higher PSNR and
SSIM on both images. H&E-1 has 30.85 dB PSNR at 25% keeping. The high ratio of framework is providing small
non-negative vectors. The image and the laboratory values shows in the table are showing different values for
different experiments. We basically using the similar dictionary-based analysis. Both can use a feature key, a
value, an optional mask, a logged update rule, and a sparse output. It is not biological coherence. FRQI, NEQR,
and QPIXL provide quantum image representations. Q-MedDict is a classical dictionary simulation inspired by
gate actions.
8. Limitations
The main limitations are substantial.
1) the data about image and lab-based data are not from
the same patients due to lack of patients data.
2) The lab-based data utilizes the metadata.
3) The image sources only two web-sourced examples.
9. Future work
In our next research, we will focus on demographic lung cancer data from known sources. All the experiments,
lab data, and public data will be verified before going to the experiments. Moreover, we will compare with
classical machine leaning outcomes with proposed quantum-based outcomes.
10. Conclusion
Our proposed method Q-MedDict is a classical quantum-based dictionary approach. It presents data with
respect to bit-string keys and applies gate-like values. It is a simple method to utilize quantum concept without
using quantum hardware. With this approach, we can achieve real quantum level outcomes using classical
computer. We have used multiple data such DCT and PCA/KLT images. DCT outperforms the Hadamard gate on
both demonstration images more than 25%. However, quantum-based dictionary model provides reliable
outcomes and framework.
CRediT Author Contribution Statement
Sazia Parvin: Conceptualization, Formal analysis, Methodology, Software, Validation. Md Sarwar Kamal: Data
curation, Visualization, Writing - Original draft, Supervision, Writing - Reviewing and editing. All authors have
read and agreed to the published version of the manuscript.
Ethics Statement
This work uses public, de-identified data documentation and public web images. No new participant
recruitment, intervention, or private health record access was performed.
Funding Declaration
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-
profit sectors.
Data Availability Statement
Lung images data are used from the publicly available sources. The link is given bellow:
https://portal.imaging.datacommons.cancer.gov/collections/cmb_lca/.
Laboratory metadata are available from the following sources:
https://wwwn.cdc.gov/Nchs/Data/Nhanes/Public/2021/DataFiles/CBC_L.htm,
https://wwwn.cdc.gov/Nchs/Data/Nhanes/Public/2021/DataFiles/BIOPRO_L.htm and
https://wwwn.cdc.gov/Nchs/Data/Nhanes/Public/2021/DataFiles/ALB_CR_L.htm.
Conflict of Interest
The authors received no specific financial support for the research, authorship, or publication of this work.
Artificial Intelligence (AI) Use Disclosure
The authors declare that Generative AI tools were used for language editing and manuscript revision support.
The authors reviewed the scientific content and checked the numerical results.
Supporting Information
Not applicable.
References
[1]
M. S. Kamal, S. F. Nimmy, Quantum advantages with dictionary-based quantum data centre design for
medical data, In: S. Roy Chowdhury, N. Dey, T. Guha, S. J. Fong, (eds) Proceedings of International
Symposium on Innovations in Computer Vision and Applications (ISICVA 2025) Lecture Notes in
Networks and Systems, Springer, Cham., 2026, 2015, doi: 10.1007/978-3-032-28206-4_2.
[2]
L. Jacob, K. Thomas, L. C. Samuel, S. S. Thomas, Chapter 11 - Fusion Techniques for medical imaging and
clinical data towards precision diagnostics and personalized care, Ed(s): S. Eslamian, P. Nanjundan, J.
George, F. Eslamian, Multimodal Learning Using Heterogeneous Data, Morgan Kaufmann, 2026, 139-
155, doi; 10.1016/B978-0-443-27528-9.00010-3.
[3]
A. Siva M, C. L. Chowdhary Multi-modal personalized federated learning with adaptive differential
privacy for medical image classification and a privacy-preserving approach, Scientific Reports, 2026, 16,
21743, 10.1038/s41598-026-49896-z.
[4]
N. Liu, G. Han, Q. Gu, Y. Zhang, M. Chen, A new era of precision diagnosis and treatment for lung cancer:
Artificial intelligence-driven multimodal data integration and clinical applications, Cell Death & Disease,
2026, 17, 534, doi: 10.1038/s41419-026-08769-z.
[5]
F. Ahmed, N. S. Naz, S. Khan, A. U. Rehman, W. M. Ismael, M. A. Khan, Explainable artificial intelligence
(xai) in medical imaging: a systematic review of techniques, applications, and challenges, BMC Medical
Imaging, 2026, 26, 37, doi: 10.1186/s12880-025-02118-w.
[6]
R. Vavekanand, A comprehensive review of multimodal large language models for medical imaging and
omics data, Archives of Computational Methods in Engineering, 2026, 33, 6149–6171, doi:
10.1007/s11831-026-10504-y.
[7]
M. Liu, J. Villazon, A. Forjaz, X. Tian, R. Fan, S. Wang, L. Shi, D. Wirtz, A. L. Kiemen, 3D multi-omics tumour
atlases: from technology to biology and clinical translation, Nature Reviews Cancer, 2026, 26, 762–789,
doi: 10.1038/s41568-026-00940-0.
[8]
Z. Li, Y. Li, J. Xiang, X. Wang, S. Yang, X. Zhang, F. Eweje, Y. Chen, X. Luo, Y. Li, J. Mulholland, C. Bergstrom,
T. Kim, F. Maria Olguin, S. Willens, J. J. Nirschl, R. West, J. Neal, M. Diehn, R. Li , AI-enabled virtual spatial
proteomics from histopathology for interpretable biomarker discovery in lung cancer, Nature Medicine,
2026, 32, 231–244, 2026, doi: 10.1038/s41591-025-04060-4.
[9]
K. S. Rawat, M. Yadav, Analyzing quantum computing applications across key scientific domains using
trends and visual analytics, Archives of Computational Methods in Engineering, 2026, 33, 401–432, doi:
10.1007/s11831-025-10312-w.
[10]
A. Sar, A. Kumar, S. Roy, A. Kaushish, T. Choudhury, S. S. Alqahtani, A. Abraham, Quantum machine
learning in medical image analysis: From diagnostics to surgery planning, IEEE Access, 2026, 14, 27690-
27732, 2026, doi: 10.1109/ACCESS.2026.3663498.
[11]
P. Suyal, K. K. Gola, C. Chakraborty, R. Kanauzia, M. Suyal, Mridula, Applications of quantum-inspired soft
computing for intelligent data processing in real-life scenarios-Chapter 10, in Quantum-Inspired
Approaches for Intelligent Data Processing, Editor(s): B. Balusamy, S. Avdhesh Yadav, S. Ramesh, M.
Vinoth Kumar, 223–257, 2026, doi: 10.1002/9781394336449.ch10.
[12]
C. Daka, S. Bhattacharyya, A novel quantum convolutional neural network framework for quantum
enhanced classification of pixelated colour images, Scientific Reports, 2026, 16, 10828, doi:
10.1038/s41598-026-45140-w.
[13]
S. Somvanshi, M. M. Islam, A. Rafe, A. G. Tusti, A. Chakraborty, A. Baitullah, T. I. Chowdhury, N. Alnawmasi,
A. Dutta, S. Das, Bridging the black box: A survey on mechanistic interpretability in AI, ACM Computing
Surveys, 2026, 58, 1–35, doi: 10.1145/3787104.
[14]
Q. Wan, C. Gao, R. Wang, X. Chen, A survey on interpretability in visual recognition, IEEE Transactions
on Pattern Analysis and Machine Intelligence, 2026, 48, 8547–8566, doi: 10.1109/TPAMI.2026.3672629.
[15]
A. Sadeghi, F. Hajati, A. Argha, N. H. Lovell, M. Yang, H. Alinejad-Rokny, Interpretable graph-based models
on multimodal biomedical data integration: A technical review and benchmarking, Nature
Communications, 2026, 17, 5405, doi: 10.1038/s41467-026-74126-5.
[16]
A. Muneer, M. Waqas, M. B. Saad, E. Showkatian, R. Bandyopadhyay, H. Xu, W. Li, J. Y. Chang, Z. Liao, C.
Haymaker, L. S. Soto, C. C. Wu, N. I. Vokes, X. Le, L. A. Byers, D. L. Gibbons, J. V. Heymach, J. Zhang, J. Wu,
From classical machine learning to emerging foundation models: Review on multimodal data integration
for cancer research, Artificial Intelligence Review, 2026, 59, 119, doi: 10.1007/s10462-026-11522-9.
[17]
X. Luo, X. Wang, F. Eweje, X. Zhang, J. L. G. Marti, S. Cascarino, S. Yang, Y. Li, R. Quinton, J. Xiang, Y. Ji, Z. Li,
Y. Chen, C. Bergstrom, T. Kim, F. M. Olguin, K. Yuan, M. Abikenari, A. Heider, S. Willens, R. Li, Ensemble
learning of pathology foundation models for precision oncology, Cancer Cell, 2026, 44, 1981–1996, doi:
10.1016/j.ccell.2026.08.008.
[18]
S. Rostami, C. Gue´rendel, M. Soliman, H. W. Stutterheim, O. Maxouri, D. I. Rodr´ıguez Sa´nchez, S.
Ursprung, N. Boveradze, G. Agrotis, K. Chupetlovska, F. Castagnoli, F. Landolfi. E. K. Hong. A. D. Pizzi, N.
Gennaro, M. A. Abdelatty, W. Jutidamrongphan, L. Petrychenko, P. Matkulcik, A. Salgado-Parente, F. M.
Arico, S. Benson, P. Snaebjornsson, Z. Bodalal, R. G. H. Beets-Tan, Tumor morphology on ct radiomics is
largely driven by the local anatomical environment, not the primary tumor type, European Radiology
Experimental, 2026, 10, 10.1186/s41747-026-00691-5.
[19]
B. S. Sahin, M. S. Khalil, A transparent rule-based screening and personalized nutrition recommendation
framework for biochemical health data: Development and validation in 46,750 NHANES adults, Journal
of Intelligent Decision Making and Information Science, 2026, 3, 2598–2616.
[20]
S. Herna´ndez-Orozco, A. Uthamacumaran, F. Herna´ndez Quiroz, K. Saeb-Parsy, H. Zenil, Integrative
adaptive indexes from noisy routine haematological markers can predict and discriminate health status
and biological age, Computers in Biology and Medicine, 2026, 208, 11162, doi:
10.1016/j.compbiomed.2026.111628.
[21]
B. Suwatronnakorn, A. Wipattanakitcharoen, T. Wuttiputhanun, A. Leelahavanichkul, N. Townamchai, Y.
Avihingsanon, S. Udomkarnjananun. Reliability of spot urine protein-to-creatinine and urine albumin-
to-creatinine ratios in acute kidney injury: A cohort study, American Journal of Kidney Diseases, 2026,
doi: 10.1053/j.ajkd.2026.06.026.
[22]
H. Zhang, Y. Wang, J. Wang, H. Li, Interpreting the black box: Interpretable machine learning and systems
pharmacology in small-molecule therapeutics, Pharmaceutics, 2026, 18, 743, doi:
10.3390/pharmaceutics18060743.
[23]
J. Spronck, L. van Eekelen, D. van Midden, J. Bogaerts, L. Tessier, V. Dechering, M. Demirel-Andishmand,
G. S. de Souza, R. Nemeth, E. Munari, G. Bogina, I. Girolami, A. Eccher, B. Acs, C. Boyaci, N. Klubickova, M.
Looijen-Salamon, S. Vos, F. Ciompi, A tissue and cell-level annotated H&E and PD-L1 histopathology
image dataset in non-small cell lung cancer, IEEE Journal of Biomedical and Health Informatics, 2026, 1–
11, doi: 10.1109/JBHI.2026.3685529.
[24]
J. Monkman, A. Kilgallon, C. Lawler, R. Tubelleza, T. N. Aung, J. H. Warrell, I. Vathiotis, I. P. Trontzas, N.
Gavrielatou, N. N. N. Chan, R. Czertok, S. Bookstein, K. O’Byrne, E. Markovits, D. L. Rimm, A. Kulasinghe,
Metabolic characterization of tumor-immune interactions by multiplexed immunofluorescence reveals
spatial mechanisms of immunotherapy response in non-small cell lung carcinoma (NSCLC), Nature
Communications, 2026, 17, 837, doi: 10.1038/s41467-026-68633-8.
[25]
X. Liu, Y. Liu, W. Duan, J. Weng, H. Dai, F. Liu, L. Wang, M. Qian, B. Zhu, S. Wu, Y. Ye, M. Yuan, S. Hong, M.
Wang, Y. Cheng, T. Chen, D. Wu, Y. Song, X. Wang, Single- cell identifies and validates human circulating
Treg subtype/state Tregfci in non-small cell lung cancer, Signal Transduction and Targeted Therapy,
2026, 11, 211, 2026, doi: 10.1038/s41392-026-02677-6.
[26]
M. Campisi, T. Osaki, I. Dryg, C. Stornante, J. Wolff, J. Weirather, N. Weaver, M. Tarannum, I. Gillanders, A.
Bers, E. Bobilev, M. S. Lineberry, P. Schol, M. Chen, K. Ota, S. R. Park, C. G. Fahey, T. C. Thai, Y. Li, Z. Li, Z-H.
Li, S. Shelton, S. Hirose, C. Padrick, E, Ivanova, M. Ha, S. Yang, H. Olszewski, S. Kivlehan, P. Lizotte, K. Hagen,
I. Gjeci, W. Haller, L. M. Zasadil, T. El Zarif, J. E. Berchuck, E. H. Knelson, E. J. Brea, I. Odintsov, T. Liu, V.
Chiono, B. Tuladhar, , K. Ngo, E. Rosentrater, J. Raizer, N. Lineberry, M. Sur, K. Pfaff, M. Freedman, F. S.
Hodi, M. Y. Tolstorukov, M. G. Oser, M. Sheffer, C. S. Mitsiades, M. J. Barnes, V. A. Appleman, P. C. Gokhale,
R. D. Kamm, C. P. Paweletz, R. Romee, S. Rodig, D. A. Barbie, N. R. Mahadevan, Vascular STING activation
facilitates NK cell anti-tumor immunity in small cell lung cancer, Cancer Cell, 2026, 44, 858–878.e16, doi:
10.1016/j.ccell.2026.02.008.
[27]
Q. Li, S. Zhang, J. Jiang, H. Li, Z. Li, J. Zhu, J. Tong, F. Wu, Research frontiers and emerging trends in sex and
gender differences in chronic obstructive pulmonary disease: A bibliometric analysis, Journal of
Thoracic Disease, 2026, doi: 10.21037/jtd-2025-1988.
[28]
N. A. Alwan, N. M. Al-Saidi, S. J. Obaiys, N. F. B. M. Noor, M. Ahmad, A comprehensive review on
classification of quantum image representations, Archives of Computational Methods in Engineering,
2026, 33, 3123–3181, doi: 10.1007/s11831-025-10369-7.
[29]
T. S. Atia, Z. A. Ali, A. Y. Yousuf, Soft-quantum computing integration: automated design and optimization
of scrambling circuits for NEQR and GNEQR image using GAs and fuzzy logic, Quantum Information
Processing, 25, 246, doi: 10.1007/s11128-026-05256-4.
[30]
N. Vinayak, S. Ahmad, Qxrnet: a hybrid CNN–QNN model with resolution-conditioned feature extraction
and variational quantum circuit, Quantum Machine Intelligence, 2026, 8, 2026, doi: 10.1007/s42484-
026-00377-6.
[31]
A. H. Salehi Shayegan, Quantum image processing filters based on FRQI representation, Quantum
Information Processing, 2026, 25, 293, doi: 10.1007/s11128-026-05321-y.
[32]
N. Tarek, An optimal hybrid quantum-classical representation for robust photonic image processing on
NISQ devices, Quantum Information Processing, 2026, 25, 295, doi: 10.1007/s11128-026-05314-x.
[33]
A. Xu, B. Sun, Z. Wang, S.-M. Fei, Separability criteria via type-ii discrete cosine transform, Advanced
Quantum Technologies, 2026, 9, e70453, doi: 10.1002/qute.70453.
Publisher Note: The views, statements, and data in all publications solely belong to the authors and
contributors. GR Scholastic is not responsible for any injury resulting from the ideas, methods, or products
mentioned. GR Scholastic remains neutral regarding jurisdictional claims in published maps and institutional
affiliations.
Open Access
This article is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License, which
permits the non-commercial use, sharing, adaptation, distribution and reproduction in any medium or format,
as long as appropriate credit to the original author(s) and the source is given by providing a link to the Creative
Commons License and changes need to be indicated if there are any. The images or other third-party material
in this article are included in the article's Creative Commons License, unless indicated otherwise in a credit line
to the material. If material is not included in the article's Creative Commons License and your intended use is
not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly
from the copyright holder. To view a copy of this License, visit: https://creativecommons.org/licenses/by-
nc/4.0/
© The Author(s) 2026
