Page 37 - GTM-2-3

Global Translational Medicine Deep learning by NMR-biochemical

done by trained operators on selected brain tissue voxels or imaging, and genomic data, toward patient-specific
brain regions . outcomes for precision medicine. Adapting the “Precision
[17]
Medicine Initiative” includes current radiological
2. Clinical disorders and MRSI applications interpretation from “average patient” to interpret the
toward DL precise patient-care management decision specific to an

The MRSI is now used in the clinic to help diagnose, treat, individual patient.
and follow cancer and lesion progression in patients. DL 3. Examples of MRI- and MRS-based DL
of MRSI data visualizes metabolite peaks as spectromics
fingerprints (relaxation constants+metabolite ratio or Three examples of breast tumor, glioblastoma (GBM),
concentrations) from different diseased brain regions, as and MS lesions are illustrated toward understanding the
shown in Figures 1 and 2. value of DL as a potential “Precision Medicine Initiative” to
evaluate disease progression, correlation with a disability,
2.1. Segmentation, coregistration, validation, and prognosis in occult diseases, correlation with fatigue, and
feature analysis by artificial intelligence in vivo theranostic effect of therapies.
Artificial intelligence‑related entities, such as Kernel 3.1. Classification of breast tumors to determine
Null Foley‑Sammon transform (KNFST), support reoccurrence and future risk
vector machines, third (PC3) and fourth (PC4) degree
polynomial, fuzzy c-means (FCM) clustering algorithm, The breast image analysis using CAD predicts the breast
K‑nearest neighbors, CNN, regression, the pattern of lesion characteristics on clinical images. The CAD is useful
the trained segmentation data set, image registration, in performing clinical tasks, such as risk assessment,
pixel coordinates, delineation, and localized NMR peak detection, screening, diagnosis, theranostic response,
characteristics, can be used to extract the tissue pathology recurrence, and others, using “virtual digital biopsies,” as
or disease burden or tumor stage using color-coding and shown in Figures 4 and 5 .
[18]
biophysical signal measurements . However, all these The DL of MRSI interpretation explores digital metabolic
[18]
methods need optimization or trimming of the input data lesion segmentation-based features of lesion volume,
by one or other methods. Thus, the maximum outcome of sphericity, texture and uptake, maximum enhancement
tissue disease burden close to real disease-specific etiology variance (dynamic contract-enhanced [DCE]), margin
or pathology obtained from biopsies can be measured sharpness on ImageNet and texture analysis T2-weighted
by combining artificial intelligence outcomes from [T2wt] MRI and apparent diffusion coefficient (diffusion-
multimodal signal processing methods. weighted imaging [DWI])] with minimum computer data
2.2. Spectromics fingerprints in precision medicine processing, as shown in Figures 6 and 7.
by multimodal MRSI with physiological screening The DL feature extraction on CADx tasks minimizes
DL by integrating MRI with multiparametric modalities extensive computing, as shown in Figure 8.
answers metabolic spectral distribution. The author The artificial intelligence neural network-analyzed CNNs-
proposes spectromics fingerprint to define tissue based breast image features classify the malignant and
metabolite distribution as the fingerprint of disease burden benign breast tumors using mammography, ultrasound,
by DL. Spectromics is an integration of multiple clinical and DCE-MRI methods as visual likelihoods of malignant
patient data, molecular imaging, and genomics modalities lesions with DL performance, as shown in Figure 9.
toward the “Precision Medicine Initiative,” as shown in
Table 2. It is a precise interpretation to decide on a patient- 3.1.1. Computerized image-based breast cancer
care management plan specific to the individual patients, recurrence risk measurement
as shown in Figures 2-4. Breast cancer risk or density measurement assessed by
Now, image analysis by computer-aided diagnosis DCE-MRI and mammography screen out high-risk
(CAD) using quantitative breast tumor metabolites and women. Cancer-risk spectromic features on DCE-MRI
NMR relaxation constants and MRI image analysis can are breast metabolite density, parenchyma texture pattern
reflect the associations of clinical, pathologic, and genomic on digital mammograms, and background parenchymal
data (genomic measurements) in tumor phenotypes. For enhancement (BPE) signal. The MRI‑visible breast
example, brain tumors, multiple sclerosis (MS) lesions, volume-growing algorithm classifies the fibroglandular and
Alzheimer’s disease, epilepsy episodes, and effective MRS-visible fat depots. Enhanced breast fibroglandular
cancer theranosis rely on the combined information kinetic curves in breast regions categorize the BPE signal
from multiple patient tests, including molecular, clinical, using FCM clustering. The mammogram images show a

Volume 2 Issue 3 (2023) 5 https://doi.org/10.36922/gtm.337