Genetic and phenotypic variation are observed between tumors of different tissue and cell types which mainly due to tumor heterogeneity. Heterogeneity is therefore as an important feature of malignant tumor in diverse genetic backgrounds, pathological patterns, differentiation stages, genetic mutation spectrum, transcriptomics and proteomics gene expression profile, et al. It indicates the high complexity and diversity in cancer progression. Tumor heterogeneity and drug resistance are great challenges for precision oncology and are taken account for the process of research and development of new diagnostic methods and further antitumor agents. Tumor spatial heterogeneity is an important prognostic factor, which may be reflected in medical images. Image texture analysis is an approach to quantifying heterogeneity that may not be appreciated by the naked eye. Different methods can be applied including statistical-, model-, and transform-based methods. Many techniques are proposed for the quantification of tumor heterogeneity as an imaging biomarker for differentiation between tumor types, tumor grading, and response monitoring and outcome prediction. In this review, the advancement in the field of tumor heterogeneity quantification and analysis with medical images were reviewed. Modern medical imaging techniques, such as magnetic resonance imaging (MRI), positron emission tomography (PET), computed tomography (CT) and ultrasound (US) in quantifying the tumor heterogeneity were summarized, and clinical studies to correlate the features with tumor responses to radiochemotherapy and prognosis value were also analyzed．
Tumor heterogeneity; Genomic; Microenvironment; Visualize; Molecular imaging; Clinical outcome
The popular application of targeted anti-tumor agents has greatly improved the efficacies of tumor therapy. However, some patients develop drug resistance after the administration and finally lead to the failure of treatment which mainly due to tumor heterogeneity. Over the last years, genomic studies also have demonstrated the complex and heterogeneous landscape in cancer [1-2]. Genetic and phenotypic variation is observed between tumors of different tissues and cell types. Human cancers frequently display substantial tumor heterogeneity in virtually all distinguishable phenotypic features, not only for cellular morphology, gene expression profile of growth factors and tumor markers, also for some metabolism characteristics, proliferative and metastatic potency, et al [1, 3]. It is moreover widely accepted that extensive genetic and phenotypic variation exists not only between tumors which are inter-tumor heterogeneity but also within individual tumors which are called intratumoral heterogeneity. In tumors, populations of genetically distinct sub-clones can intermingle or be spatially separated, and this sub-clonal architecture varies dynamically throughout the disease course . Inter- and intratumoral heterogeneity have significant implications for the choice of biomarkers to guide clinical decision making in cancer medicine. Such tumor heterogeneity has revealed its potential implications for treatment response and prognosis . The relevance of subclonal mutations to cancer outcome and therapeutic response requires further investigation and review
During the last decades, innovations in medical imaging make the detection of tumor heterogeneity to move towards the era of quantitative analysis –radiomics and radio genomics .The innovation of medical equipment mainly includes the improvement of image hardware and the development of combined modality machines. The combination of PET and CT achieve the perfect combination of organizational structure and functional metabolism evaluation. The application of them makes it possible that imaging technologies can evaluate whole tumor heterogeneity by capturing images information in the good spatial resolution . The methods which can be applied in quantifying tumor heterogeneity includes selective of region of interest (ROI), histogram analysis, texture analysis, spatial geometric approaches and clustering analysis. Among these methods, texture analysis can evaluate the gray-level intensity and decode spatial correction between voxels within an image so that to derive so-called ‘texture features’ which can provide the measure of intraregional heterogeneity [7-9]. Tumor-heterogeneity-based imaging parameters provide essential information of tumor biological/functional properties that can help the diagnosis of cancers and influence long-term treatment outcome and predict the response to therapy .
In this review, we summarize the evidence for genetic and phenotypic variation both within and between tumors, and the advancement in the field of tumor heterogeneity quantification and analysis with molecular imaging were also reviewed. Specifically, the review focused on the role of genomic instability in generating inter- and intratumoral heterogeneity and discussed how tumor heterogeneity might affect tumor biology, drug response and patient outcome. Modern medical imaging approaches in cancer diagnosis based on the tumor heterogeneity, such as ROI, histogram analysis, texture analysis, spatial geometric approaches and clustering analysis in quantifying the tumor heterogeneity were summarized，and clinical studies to correlate the features with tumor responses to radiochemotherapy and prognosis value was also analyzed.
HETEROGENEITY BETWEEN DIFFERENT TUMORS AND DIVERSITY WITHIN INDIVIDUAL TUMORS
The tumor for a patient is often a mixture of multiple genotypically and phenotypically distinct cell populations and therefore it contributes to failures of targeted therapies and to drug resistance. Tumors that originate from different tissues and cell types vary in terms of their genomic landscapes, prognosis and their response to cytotoxic therapies, probably resulting from the fact that the genetic events of transformation interact with cell-intrinsic biological properties. Reflecting this, the site of origin of the primary tumor frequently determines treatment decisions. However, considerable variation in terms of genomic aberrations, aggressiveness, and drug sensitivity is also observed between tumors that originate from the same tissue and cell type [11-12] .
Mutational frequencies of oncogenes and tumor suppressors vary between tumors of different tissues, probably reflecting the importance of distinct signaling pathways within specific tissues or cellular contexts . The clinical challenge that intratumor variation presents can be met to some extent by classifying tumors into subgroups. These may predict the patient outcome or drug sensitivity on the basis of mutations, copy number variation (CNV), protein or RNA expression profiles, or patterns of genomic instability . Recent advances in next-generation sequencing have revealed greater than expected inter- and intratumoral genetic heterogeneity. The results showed that very few mutations were observed in more than 5-10% of tumors of a particular tissue type . Moreover, the same gene within a cohort of tumors may be affected by some point mutation, CNV, epigenetics or the combination of these, which underscores the need for integrative approaches to analyzing somatic aberrations in the cancer genome.
There are two models for explaining the heterogeneity of intratumor. One is about the random process that would result in no specific distribution, and the other one is a selective process that may lead to identifiable compartments including intraepithelial components. The cancer-associated genetic instability would induce independent evolution in different tumor areas independent of the intraepithelial or invasive location. High-grade pathologies are the most reliably diagnosed lesion and progress with morphology and molecular alterations which involved the acquisition of invasion and metastasis capability. Such changes could occur before the morphological evidence of malignancy and they also reflect the intratumoral heterogeneity of metastasis in cytologically low-grade pathologies (multiple parallel pathways) that not necessarily promote via the high-grade stage (linear pathways). In contrast to linear models of tumor evolution with sequentially ordered somatic mutations in driver genes resulting in clonal sweeps of homogeneous tumor cell expansion, the new evidence suggests that branched evolutionary tumor growth may contribute to the genetic heterogeneity of intratumor both within primary tumors and between primary and metastatic sites of tumor.
It has been well established of the impact of heterogeneity on tumor growth control from an evolutionary perspective . The clonal evolution model of cancer was proposed and with this evolutionary model not only to explain the tumor growth and treatment failure but the increased tumor metastasis that occurs during the natural history of advanced solid tumors. The new microenvironment, together with the selection pressures imposed by the process of metastasis itself at metastatic, offers a plausible explanation for genetic differences between metastatic and primary sites. However, it is unclear precisely how spatial separation of genetically distinct clones arises in primary tumors . It is possible that the separation reflects the presence of distinct microenvironmental niches in the primary tumor, such that the subclones occupying each niche evolve relatively independently of one another.
Tumor heterogeneity is driven by the introduction of genetic and/or epigenetic alterations which induced by genomic instability, and the evolutionary selection . Though evolution is driven by the selection of phenotypes according to their relative fitness, not all somatic genetic alterations could induce recognizable phenotypic consequence, and even fewer provide a fitness advantage. In the study of cancer evolution, it is better to combine functional screening with multidimensional phenotyping including signaling pathway, epigenetic modification, transcriptional, metabolic as well as other alterations along with genetic alterations. All of these will be most informative in revealing the sources of the phenotypes driving tumorigenesis. Regarding the contribution of heterogeneity, although heterogeneity can be broadly considered to be a trait that allows tumors to overcome evolutionary pressures, it can also exploit therapeutically for future. This, therefore, makes it important to develop some tools to quantify and model the heterogeneity of tumor.
HETEROGENEITY, MICROENVIRONMENT, AND METASTASIS
Cancer is a systemic disease and advanced cancers can be locally advanced or metastatic. Generally, the metastatic potential is through to be due to tumor heterogeneities which origin from the differences generated in the evolution process of cancer itself as well as host microenvironment. The capability of invasion and metastasis is closely related to cell motility and requires the cytoskeleton that is essential during mitoses . Since malignancy criteria are mainly related to the phenotype of actively proliferating cells, metastatic deposits genetically match well-differentiated areas of primary and invasive areas . These factors need attention when planning the evaluation of intratumoral heterogeneity and detailed specification would be necessary to pinpoint the intratumoral location, to assess selective process and to analyze pathways [18, 20]. These biological foundations would improve the therapeutic design and patient’s management base on heterogeneity.
The connection between initial stages and metastasis of cancer comes from tumor heterogeneity and progression features. Progression of neoplasm is the acquired cell growth capability in surrounding or distant tissues, it also means the acquisition of invasive capacities for intraepithelial pathologies and metastatic capacities . This capability reflects the interaction between tumor cells and the microenvironment. It has been well established that cancer cells can survive and proliferate only at specific secondary sites where releases molecular mediators suitable for that type of cancer cells [21-23]. Metastasis, however, is a rather inefficient process. Its formation is a complex process that requires tumor cells to escape from primary site and intrastate into the hematic or lymphatic circulation, then migrate and extravasate further into the secondary organs . In order to form metastases, primary tumors would produce factors and induce the formation of the compatible environment in the organs where metastasis might be seeded. This environment was called niche by which a special, permissive microenvironment in secondary target organs is induced by the primary tumor .
Studies that compared the genetic composition of primary tumors and secondary metastatic sites has been revealed the close clonal relationships between the two in the majority of cases . Moreover, primary tumors also showed similar gene expression patterns with the metastatic sites . Another prediction from the linear model of tumor progression is that different metastases displayed close clonal relationships among each other and different metastatic pathologies within the same patients demonstrated similar clonal relationships, signifying monoclonal origin . While the evidence of the close genetic relationship between primary and metastatic tumors is compelling, some cases display dramatic divergence, challenging the model where the acquisition of metastasis is thought to be the final step of progression. Important clues about the clonal evolution of tumors and the relationship between primary tumors and metastatic sites can be gained by the analysis of disseminated tumor cells and some studies were suggested the importance of relatively early metastatic spread .
Currently, the therapeutic decisions of tumors are based on the analysis of primary tumor specimens. However, it should be noted that this approach could only be reliable only where the genetic compositions of primary and metastatic tumors are similar. In addition, elucidating the situation of early versus late origin of metastasis-initiating cells is pivotal for determining the therapeutic approaches that target tumor cell invasion. The question of clonal heterogeneity within metastases remains unexplored, and, notably, clonal heterogeneity within primary tumors has limited direct therapeutic relevance, as in most cases primary tumors can be successfully removed by surgery, but this is most often not the case for metastatic outgrowths. Therefore, elimination of metastatic clones has to rely on adjuvant therapies, and it is likely that the success of these therapies will depend on the extent of the heterogeneity of these tumor cell populations.
VISUALIZE TUMOR’S HETEROGENEITY BY MOLECULAR IMAGING
As we reviewed in the previous part, the heterogeneity involves variations in genomic subtypes, the expression profile of growth factors , as well as the tumoral microenvironment, which lead to regional differences within individual tumors in proliferation, cell death, metabolic activity, vascularity [30-31]. However, tumor heterogeneity is challenging to characterize and quantify in the clinical setting. Tissue biopsies are invasive and provide only limited sampling points of the entire heterogeneous tumor volume. Molecular imaging modalities, such as PET, MRI, CT and US, have the potential to provide whole-lesion information noninvasively. PET shows great benefits in staging and therapy response evaluation of lung cancer and lymphoma  MRI, including functional and spectroscopic techniques showed great potential in imaging intratumoral heterogeneity and identifying biological subpopulations within single tumors, distinguishing for example viable tumor from necrosis and adipose healthy tissues , or classifying the tumor microenvironment in terms of vascular heterogeneity, hypoxia, and acidity . Imaging technologies, as minimally invasive or non-invasive and also repeatedly, can evaluate whole tumor heterogeneity by capturing images information in the good spatial resolution .
Heterogeneity of tumor microenvironment has been reported with respect to regional vascular density and hypoxia, proliferation, energy metabolism, and gene expression of tumors . In MR, ADC value has been considered an inverse index of tumor cellularity, as high cellularity is associated with relative reductions in extracellular space, resulting in decreased diffusivity of water molecules .
Visualizing tumor metabolic heterogeneity Metabolic reprogramming, fluctuations in bioenergetic fuels and modulation of oxidative stress became new key hallmarks of tumor development. In cancer, elevated glucose uptake and high glycolytic rate constitute a growth advantage for tumors. This is the basis of the universally known Warburg effect, the gold standard process in cancer metabolism. It gave rise to one major clinical application for detecting cancer cells using glucose analogs: the PET scan imaging. With different genetic variants in the process of growth and treatment, tumor cells will also exhibit obvious heterogeneity in glucose metabolic characteristics . Different with normal cell’s metabolism, the tumor cells decompose glucose to provide energy for aerobic fermentation by Warburg effect and accelerate the absorption and utilization of glucose so that provide sufficient energy for the rapid growth and proliferation of tumor cells . Although aerobic glycolysis is the main type of energy metabolism of tumor cells, in fact, even in dependence on aerobic glycolysis in cancer cells, the oxidative phosphorylation did not lack oxidative phosphorylation (OXPHOS) has not completely stopped, some cells even depend on oxidative phosphorylation for energy . Also, there are two kinds of glucose metabolism patterns coexisted in the same tumor subtype, and the metabolic patterns of the tumor and non-tumor stem are different. As the most abundant free amino acid in the human blood, glutamine catabolism is another essential metabolic feature of tumor cells. The basic reasons of the metabolic heterogeneity that cancer is a genetic disease . The heterogeneity of the genetic background determines the diversity of tumor metabolism, which will lead to different tumor phenotypes and respect to treatment.
Currently, the PET is the most common technique used in cancer metabolic imaging which as one crucial aspect in many different types of cancers’ diagnoses and treatment response monitoring. Besides, PET is usually combined with diagnostic CT scan and the radiotracer FDG which exploits the expanded glucose consumption of cancer cells to image tumors in a non-invasive manner. In a study of a large cohort of 234 sarcoma patients with PET scanning before neoadjuvant chemotherapy or surgical resection, revealed that tumors with highly heterogeneous metabolism likely have a high metastatic potential associated with poor patient outcome . While in another study, adenocarcinomas of lung cancer are proved to be more metabolically heterogeneous than epidermoid tumors .
Visualizing vascular tumor heterogeneity The proliferation of tumor cells is faster than angiogenesis during tumor growth, which leadsto the formation of poorly perfused hypoxic zones near the tumor center and finally leads to the heterogeneity of vascularity . Vascular heterogeneity may cause localized reductions in blood flow, increased risk of invasion and metastasis, impaired delivery of chemotherapeutic agents, increased cellular resistance to drug therapy, as well as radiotherapy and inhibition of immune responses. Experimental evidence has shown that vascular heterogeneity can be associated with worse disease progression, poor therapeutic response and malignancy. The possible mechanism is connected with the selection of aggressive tumor cells that can tolerate the hostile acidic, hypoxic environment in poorly perfused low-enhancing regions. Tumor vascular heterogeneity alteration induced by treatment has been reported in several preclinical and clinical settings .
As is mentioned above, tumor microvasculature possesses a high degree of heterogeneity in its structure and function. Several molecular and functional, such as DCE-MRI, PET and CT were developed to this challenge by monitoring pathophysiologic changes in various aspects of tumor vascular structure and functionality. Imaging techniques such as DCE-MRI is a powerful technique for the detection of subtle changes in tumor vascular function . The analysis of heterogeneity on DCE-MRI series in adnexal tumors is a very effective method to distinguish benign from malignant tumors, borderline from invasive tumors, and malignant tumors with from malignant tumors without carcinomatosis .
METHODS FOR QUANTIFYING CANCER HETEROGENEITY
There are several kinds of advanced analysis approaches, such as ROI, histogram analysis, texture analysis, spatial geometric approaches, and clustering analysis ,which we have already been used for the detection of tumor heterogeneity, combined with diagnostic tools like MRI, CT, PET and US. By this, it can provide a lot of valuable information for clinical[46-48].
Use of selective regions of interest (ROI) Manual selection of small regions of interest with high/low values has been used for imaging data . Many researchers prefer to focus on a small region with the highest enhancement instead of the whole tumor when it comes to some solid tumors. The ROI average is equivalent to the maximum of the whole tumor histogram if the ROI is small enough and correctly placed. The technique is simple and inter-observer variations may be of minimal importance where large variations are present. Rim enhancement patterns led to a quantitative approach to describe radial heterogeneity on MR  where the tumor is divided into concentric circles and mean or median parameter values plotted against band number. Benjaminsen et al developed this rough segmentation methodology into a quantitative approach to describe radial heterogeneity in tumor perfusion .
At the limited spatial resolution of PET images, homogeneous radiotracer uptake in an ellipsoid would appear to progressively decline away from the center and the deviation from this model has been used to quantify heterogeneity in FDG uptake and found to be a predictor of outcome in 234 sarcoma patients. That is always called model-based approach, that we have been widely used in texture analysis of tumor heterogeneity .
Histogram analysis Intratumoral heterogeneity is also reflected by the shapes of parameter distribution histograms. The intensity histogram allows a range of descriptive parameters including range, mean, median, standard deviation (SD), skewness, kurtosis, entropy and various percentile cut-off values. Disease progression is usually accompanied by an expansion of the histogram to the right-side with the peak broadening but decreased its height. While the successful treatment makes the peak narrower and more normal which lead to the histogram leftwards [51-52]. Although the heterogeneity metric characterized a rim pattern, it appeared less sensitive to heterogeneous radiotracer distribution and to a heterogeneous response than the simpler inverse of the coefficient of variation (mean/SD). Tixier et al. compared global (distribution histogram), regional (intensity–size zone matrix) and local (co-occurrence matrix) features on static FDG images for their capacity to classify 41 patients with oesophageal cancer with respect to response to chemoradiotherapy and showed that local homogeneity and entropy, as well as regional intensity and size-zone variabilities, could identify non-responders, partial and complete responders with higher sensitivity than SUV measurements . In 18F-FDG PET, Orlhac et al. found that SDHist, EntropyHist, and EnergyHist were highly correlated with SUV measurements on their patient sets so that to prove SUVs and SDHist led to a similar performance in predicting outcomes in cervical cancers and head and neck cancers . Tan et al. recognized that skewness could differentiate responding from nonresponding tumors .
Texture analysis Texture analysis 7is a way we can evaluate the gray-level intensity and position of the pixels within an image so that to derive so-called ‘texture features’ which can detect the spatial correction between voxels to better differentiate lesions with similar histogram but distinctive enhancement patterns. Several methods have already been commonly applied in this program, such as statistical-, models, and transform-based methods [8-9].The question is that different groups use different approaches, and differences in texture indices results might come from a different delineation method.
Statistical-based techniques are commonly used to describe the intensity distribution and relationships of gray-level values within an image. There are three orders of parameters to describe statistical-based texture analysis. Model-based  approaches always used on the texture analysis. Fractal analysis is a measurement of the irregularity of a surface which can also represent texture. It has been observed that the greater the fractal dimension is, the rougher the texture. Transform-based methods, such as Fourier, Gabor, and wavelet transforms, analyze texture in a frequency or the scale space.
The index we most use include energy, entropy, contrast, homogeneity, and correlation, also, the higher the entropy and the lower the energy, the higher the heterogeneity of the gray-level distribution of the tumor image. The co-occurrence matrix method most frequently used texture analysis technique in DCE-MRI data analysis, and Nie et al. further associated the textural features with the visual descriptors in the BI-RADS lexicon and give the highly mathematical approach an intuitive interpretation. Entropy, energy, contrast, correlation and homogeneity are the most reproducible MR parameter reflecting tumor heterogeneity even though manual delineation of the tumor boundary for segmentation might inflate interobserver variability compared with semi- or fully automated segmentation. Imaging-derived tumor heterogeneity can be measured directly on contrast-enhanced CT or MR images and is essential for the staging and management of lung cancer; thus, entropy may be incorporated into current practice without invasive procedures . In 18F-FDG PET for the heterogeneity analysis of NSCLC, van et al found adenocarcinomas had a lower mean value of energy and higher homogeneity than epidermoid tumors. Also, larger tumors were more heterogeneous (lower energy and higher entropy) and had lower contrast and higher correlation, that was proved by them. Although PET images suffer from modest spatial resolution, it was hypothesized that the uptake distribution within the tumor could bring more insight into the tumor than the single SUV or tumor volume [54, 60]. Many of these features have already been performed for the characterization of lesions in different organs.
Spatial geometric approaches Here individual parameters are interpreted as the height on an extruded 4D hypervolume. Geometric features like surface area, volume, surface to volume ratio and box-count fractal dimension can be calculated and used as quantitative descriptors of heterogeneity. Describing variations in both spatial and parametric domains by nature, this method can provide spatial and parametric information is treated equally as dimensions in the 4D hypervolume. Geometric features of the spatial-parametric hypervolume comprise a pool of potential biomarkers that can simultaneously quantify tumor morphology and heterogeneity .
Clinical studies show increased discriminative power between low-grade and high-grade glioma and increased ability to predict outcome in colorectal metastatic disease in response to bevacizumab . Dimitrakopoulou-Strauss et al  first used the box-counting method on time-activity curves from dynamic FDG, 15OH2O and 6-18Ffluoro-L-DOPA (as a potential radiotracer for melanin synthesis) images from 11 metastatic melanoma patients, and did not find evidence of heterogeneous distribution within the manually delineated pathologies. However, when combined with SUVs and kinetic parameters in discriminant analysis, fractal dimension improved the differentiation of sarcomas from dynamic FDG images.
Cluster analysis in images is the task of grouping a set of features in such a way that features in the same group (called a cluster) are more similar (in some sense or another) to each other than to those in other groups (clusters). It is the main task of exploratory data mining and a common technique for statistical data analysis in image analysis. Cluster analysis is attractive since it will, in theory, allow identification of tumor components with common properties. It also maintains information regarding spatial relationships of tissue subtypes.
In Metz et al study, 13 cancer patients underwent DCE-MRI for the estimation of blood volume and blood flow, static FDG PET for the measurement of glucose metabolism and static 18F-galacto-RGD (arginine-glycine-aspartate) for imaging the avb3/5 integrin receptors. After delineating the whole tumor on the fused PET images, the distribution of SUVs at the voxel level were clustered into four regions using preset thresholds for the two tracers on a scatterplot: high FDG and RGD, high FDG and low RGD, low FDG and high RGD and low FDG and RGD. Significant differences in blood volume but not blood flow were detected between regions of both high glucose metabolism and avb3/5 expression and regions of both low glucose metabolism and avb3/5 expression. No significant correlation was found between the MR and PET biomarkers when the tumors were analyzed as a whole . This method, combined with radiomics, was also used by Zhang et al in breast tumor classification.
HIGH THROUGHPUT OF HETEROGENOUS FEATURES : RADIOMICS
The physiology and anatomy of organs and tumors are driven by gene expression patterns which as a product of cellular genetics interfacing with the microenvironment. It has become clear that distinct sub-regions of tumors, identifiable by MR imaging, have distinct gene expression patterns [65-66]. Recently, there have been attempts to determine if quantitative analysis of the anatomy can be used to infer an underlying molecular gene expression pattern. This involves radiomics which is the extraction of quantitative features from radiographic images. Relating these to gene expression patterns using sophisticated bio-informatic approaches are sometimes termed radiogenomics[5-6, 67].
Radiomics refers to the extraction and analysis of a large number of quantitative features with high throughput from medical images. It’s an advanced image analysis on both conventional and novel medical “imagines, make it possible to capture additional information not currently used, and more specifically, which can help disease diagnosis, choice of the treatment decision, and evaluation of the response to chemo- or radiotherapy. Furthermore, Radiomics can be used to predict the information about gene signatures. The workflow of Radiomics include these steps as follows: (a) acquiring the images, (b) identifying the volumes of interest (ie, those that may contain prognostic value), (c) segmenting the volumes (ie, delineating the borders of the volume with computer-assisted contouring), (d) extracting and qualifying descriptive features from the volume, (e) using these to populate a searchable database, and (f) mining these data to develop classifier models to predict outcomes either alone or in combination with additional information, such as demographic, clinical, comorbidity, or genomic data.
As is known that tumors with more genomic heterogeneity are more likely to develop a resistance to treatment and metastasize. This means heterogeneous tumors have a worse prognosis. The genomic heterogeneity could translate to the heterogeneity of protein expression in tumors, which in turn leads to the changes of histopathology and can be captured through imaging, and thus provide valuable information for clinical doctors.
It has been well known that solid tumors present extraordinarily spatial and temporal heterogeneity at different levels, which will limit the use of biopsy based molecular assays but gives a huge potential for noninvasive imaging . So far, development in Medical imaging techniques such as advanced hardware, new imaging agents, and standardized protocol now allow for quantitative imaging but require the development of ‘smart’ automated software to extract more information from image-based features. Radiomics–the high-throughput extraction of image features from radiographic images – is promising to promote the precision medicine. And it has been a trend that radiomics produce promising prognostic markers when applied on large patient cohorts.
DETECTION OF HETEROGENEITY AND CLINICAL OUTCOME
It has been well know that highly metabolically active tumors with a high cellular and microvascular density are also prone to present with hypoxic areas. For the aggressiveness of carcinomas with PET/MRI, It has been shown that aggressive tumor also presented with high cellular and microvascular density, high glycolytic activity and focal tumor hypoxia . Increased tumor heterogeneity on unenhanced CT or MR images has been associated with poor prognosis in patients with cancers [46, 69-71]. Clinical treatment stratification is based on tumor biology received by minimal invasive biopsy, an accurate tissue sampling, especially in solid tumors, is highly desirable. Therefore, our noninvasive tumor classification might enable a precise target selection and biopsy guidance in solid tumors, by revealing the presence of tumor heterogeneity and detecting the most aggressive part of the tumor, and thus would reduce the risk of undergrading and under treatment . The daily dilemma in managing cancer patients is the heterogeneous therapy responsiveness among different patients with the same tumor stage and even among different subregions within the same tumor. Treatment failure is frequently not detected until many months after the completion of primary therapy. At such delayed time, salvage treatment options have limited impact on the ultimate long-term treatment outcome, primary tumor control and survival . Therefore, early prediction of failure from an ongoing treatment is the key to enable a therapeutic window to target intensified therapy to those patients with a higher risk of failure. The application of the multi parametric imaging might be the early response evaluation after one to two cycles of neoadjuvant chemotherapy (NACT) for breast cancers, in order to adapt cancer treatment regimen or to spare ineffective and toxicity-related therapy . Tumor-heterogeneity-based imaging parameters provide essential information of tumor biological/functional properties that critically influence long-term treatment outcome and predict the response to therapy .
In some locally advanced tumors, tumor hypoxia is correlated with increased resistance to hemotherapy and radiotherapy, thus diminishing the rate of local control as well as distant disease control . Traditional clinical approaches to decoding hypoxic regions are based on needle electrodes or tissue sampling. PET imaging using 18F-fluoromisonidazole (FMISO) can identify hypoxic tumor sub-volumes and track spatio-temporal dynamics. It therefore might be of considerable increased value of improved planning and monitoring of CRT for cervix cancer . In one study of cervical cancer’s heterogeneity showed that functional tumor heterogeneity can be characterized by DCE-MRI to predict ultimate long-term treatment outcome. For the prediction of therapy response of osteosarcoma (OS) with MRI , it has been demonstrated significant changes of the plasma volume fraction and vascular leakage in OS with bevacizumab. They also observed that DCE-MRI parameters were correlated with 18F-FDG PET measures of tumor metabolism during neoadjuvant therapy, and this could provide complementary information for understanding the underlying changes in tumor physiology and response to therapy .
Although the majority of spontaneous tumors derive from a single cell, people have come to realize intra-tumor heterogeneity of individual tumors. Human cancers frequently display the substantial difference in phenotypic features, such as the degree of differentiation, cell proliferation rate, invasion and metastatic potential, response to therapy and many other aspects. Molecular biology studies have confirmed the occurrence of new mutations during the process of tumor progression, which provides more powerful evidence to show the existence of intra-tumor genetic heterogeneity.
With the widespread use of imaging in clinical and research, it as a surrogate for diagnosis of solid tumors provides a fast and repeatable method that is non-invasive. Such a method provides a potential replacement for high-risk invasive biopsy procedures (which cannot always accurately reflect the biologic information of tumors because of its spatial and temporal heterogeneity) and subsequent histologic examination. Furthermore, the medical imaging innovations with updating hardware could allow for more accurate diagnosis and prognostication promotes the advancement of methods for detection of tumor heterogeneity and makes the field to move towards the era of quantitative analysis-radiomics and radiogenomics. A systematic methodology -radiomics- can be constructed by getting the image features in the multi-source data platform, combining with gene and clinical information, mining and screening the feature sets which highly correlate to the diagnosis of malignant tumors. Tumor-heterogeneity-detected by medical imaging provide essential information of tumor biological/functional properties that critically influence long-term treatment outcome and predict the response to therapy. With the methods or approaches we used for the detection of tumor heterogeneity, the information we get can not only help the diagnosis of carcinomas and distinguish malignant tumors from benign, but always used to make the decision of treatment and evaluate clinical outcome.