Dec-Feb'17 JPR

12 downloads 0 Views 2MB Size Report
breasts symmetrically. If the Hotspot is present in areas like armpit, neck and inframammary folds where the temperature is naturally expected to be more, then ...
RESEARCH PAPERS ARTICLE

HOTSPOT DETECTION IN INFRARED BREAST THERMOGRAMS FOR DISEASE IDENTIFICATION By A.T. WAKANKAR *

S. LALITHAKUMARI **

G.R. SURESH ***

* Research Scholar, Sathyabama University, Chennai, India. ** Associate Professor, Department of EIE, Sathyabama University, Chennai, India. *** Professor, Department of Electronics and Communication Engineering, Rajalakshmi College of Engineering, Chennai, India. Date Received: 30/06/2017

Date Revised: 01/07/2017

Date Accepted: 05/07/2017

ABSTRACT For ages, human body temperature has been used as an indicator for judging the health status. Over the years, the science of Medical Thermography has evolved to measure body surface temperatures that help in making relevant judgments about diseases. Unusual temperature patterns or the values above permissible limits indicate abnormality. In most of the cases, such temperature differences indicate a high chance of inflammation, infection or malignancy. Localized increase in temperature is termed as Hot spot and is indicative of such abnormalities. This paper develops an algorithm to detect the highest temperature region in breast thermogram to predict the breast disease. Thermal image is captured using infrared camera. The temperature data are processed to find the hotspot. The location and shape of the hotspot is detected. The thermograms are further segmented in left and right part manually. Statistical parameters are calculated from temperature data of segmented region. Significant differences in these parameters is observed for healthy and sick cases showing asymmetry. The presence of asymmetric hotspot suggests the further follow up. Results are validated by a radiologist confirming the performance of the algorithm. Keywords: Breast Disease, Thermography, Thermal Analysis, Hotspot Detection, Asymmetry Analysis. INTRODUCTION

tests that are very safe. Thermography, is non-invasive,

Breast cancer has been known for decades to be the

non-contact skin surface temperature screening test that

most common type of cancer among women. The

is economic, quick and does not inflict any pain on the

incidence of breast cancer in India is on the rise and is

patient [1]. Human body emits infrared rays in proportion

rapidly becoming the number one cancer in women [1].

to the body temperature. Thermal camera converts this

Breast cancer can be treated effectively only if it is

Infrared (IR) energy into electrical signals by the imaging

detected at an earlier stage. A range of imaging

sensor in the camera and displays on a monitor as a color

methods like mammography, ultrasound and MRI are

image that represents the variations of the temperature

available for breast cancer detection [3].

values [5, 9]. The examination of female breast with IR

Mammography is considered as the Gold standard

camera for diagnosing the breast cancer is based on the

i m a g i n g t e c h n o l o g y f o r b r e a s t c a n c e r. B u t,

fact that cancer tissue metabolizes more actively than

mammography is structural imaging modality that

other normal tissues and thus has a higher temperature.

exposes the breast to Xray radiations on compressed

The heat thus produced is conveyed to the skin surface

breasts. Due to high compression of breasts and

resulting in a higher temperature in the skin directly over

repeated exposure to toxic Xray mammography is not

the malignancy than in other regions [2]. Hence, the

ideal for patients [4]. Infrared Imaging is fast emerging as an option for breast

temperature distribution is altered, causing thermal asymmetry between the right and the left breasts. Thus,

No. 4 l December 2016 - February 2017 i-manager’s Journal on Pattern Recognition, Vol. 3 l

21

RESEARCH PAPERS ARTICLE the cancer cells can be imaged as hotspots in the

cases. The patients with higher LTI amplitudes have higher

infrared images [3]. In a normal healthy subject, the skin

carcinomous possibility. It determined the optimal LTI

temperature pattern is remarkably consistent showing

amplitude threshold of 1˚C for breast cancer detection

bilateral symmetry. The asymmetric pattern indicates the

from the Receiver Operating Characteristic (ROC) curve

deviation from the normal brought out by pathological

based on the rule of Youden's index maximization [6].

changes [2].

2. Methodology

The main objective of this work is to evaluate the feasibility

Breast Thermograms are captured by thermal camera and

of thermography as a breast imaging modality. Hotspot

are processed for hotspot detection. Hot spot detection

Detection Algorithm is developed in order to determine

provides the highest temperature area extraction, in order

the spots with highest temperature which indicates the

to determine the presence of lumps in the breast. The

presence of abnormality.

breast thermograms are further segmented manually in left

1. Literature Review

and right part. Statistical parameters like mean, median,

The first use of thermography was by Gorman in 1939 that

variance, skewness, kurtosis and entropy are calculated.

evaluated changes in vascular structure in breasts during

The significant difference is observed for healthy and

pregnancy with infrared photography. Lawson 1956, 57

abnormal cases. The results of algorithm are validated by a

and Amalrie 1957 initially published research regarding

radiologist. Camera software 'SMARTVIEW' is used to view

thermography and breast cancer. In 1963, Lawson and

and analyze the image.

Chughtai demonstrated empirical evidence that breast

2.1 Image Acquisition

malignancy alters regional skin surface temperature.

Infrared Camera FLUKE TiX 560 having a spectral

V. Umadevi has developed an ITBIC interpretation system

response of 8 μm to 14 μm and IR Resolution of 320X240 is

for characterization of breast thermal images. This is a tool

used for image acquisition at Magnolia Breast Care

for differentiation of breast thermal images as normal or a

Center, Pune under the guidance of Dr. Shekhar Kulkarni.

case for follow-up, which is helpful for initial and mass

Recommended acquisition protocol is followed to

screening. Statistical analysis was carried out on the

improve accuracy. The test was carried out in AC room,

results of ITBIC interpreter which resulted in Positive

which was darkened during the test and the temperature

Predictive Values (PPV) of 80%, Negative Predictive Values

is adjusted between 18 to 22 degrees Celsius. Total 20

(NPV) of 95.6%, Sensitivity of 66.7% and Specificity of

images are acquired, out of which 7 are sick and 13 are

97.7% [7]. Iman Zare, Ahmad Ghafarpour, et al described

normal cases.

that thermography is suitable compared to the

Thermal images captured by an infrared camera can be

ultrasound diagnosis in detecting the breast tissue

viewed using different color palettes. Breast Thermogram

diseases such as cystic masses and hypo echo masses

in Different Color Palates are shown in Figure 1.

by adopting asymmetr y technique. This study emphasizes that thermography should not be used for the first time diagnosis. This technology needs accurate clinical evaluation and it is likely that thermography can be a part of breast screening, detecting and etc. in near future [10]. Xianwu tang, Haishu Ding, et al described the morphological measurement of local temperature increase (LTI) amplitudes in breast infrared thermograms for breast cancer detection. By applying morphological approaches, they obtained LTI amplitudes, which have significant difference between benign and malignant 22

Figure 1(a), 1(b), 1(c) & 1 (d) is representation of the breast thermal image when viewed in different color palettes. Each breast thermogram is best viewed in a certain temperature scale and in a specific color palette. The blue red colour scale is selected in our work. Specific camera compatible software 'SMARTVIEW' is used for viewing the thermal images. Sample screenshot of SMARTVIEW is shown in Figure 2. 2.2 Sample Thermograms Figure 3 shows the thermograms of volunteers with

No. 4 l December 2016 - February 2017 i-manager’s Journal on Pattern Recognition, Vol. 3 l

RESEARCH PAPERS ARTICLE abnormal breast and with normal breast respectively. The

The thermogram is further analyzed through software by

color bar present on the thermograms gives information

selecting marker. The minimum, maximum and average

about temperature distribution indicating the coolest part

temperature of the selected area is shown in Figure 4. It

as blue and the warmest part of the as red as shown in figure

can be observed that, the left breast of volunteer is having

[8]. For abnormal cases, higher temperature indicated by

more temperature as compared to right breast. The

red color and the bilateral asymmetry can be seen.

average temperature difference of around 1˚C is observed for abnormal cases. This particular subject had undergone mammography and the lump was detected. 2.3 Hotspot Detection Each thermal image is a two dimensional matrix of size 320*240, which is the representation of breast surface

(a)

(b)

temperature. It has coordinates x and y and T(x,y) is the temperature of that pixel. The Hotspot detection algorithm consists of two parts: Body Boundary detection and extraction of highest temperature area in the thermogram. Then both the outputs are combined to create a single image which is easy to interpret [7]. The

(c) (d) Figure 1. Breast Thermogram in Different Color Palates 1a) Blue Red 1b) High Contrast 1c) Hot Metal 1d) Grayscale

flowchart of the process is as shown in Figure 5. The temperature file of the whole breast thermogram is exported as an excel file. This file is read in MATLAB which gives a matrix of 320*240 values. The minimum temperature Tmin value is found out indicating the ambient temperature. Any value less than Tmin is made zero and shown as black. While values greater than Tmin are made 1, indicating the body surface temperature. Edge detector algorithm is then applied to this new matrix to detect body contour. Figure 6(a) and 6(b) shows the output of Background and Edge detection.

Figure 2. Screenshot of SMARTVIEW

To detect the hot spot, highest value of temperature Tmax is found out from the original temperature matrix. Again these values are compared with a threshold of (Tmax-1)°C. The threshold of 1°C The threshold of 1°C is selected after

Figure 3. Sample Thermograms

Figure 4. Difference in Average Temperature

No. 4 l December 2016 - February 2017 i-manager’s Journal on Pattern Recognition, Vol. 3 l

23

RESEARCH PAPERS ARTICLE analyzing the image through SMARTVIEW where we can

generates the temperature data of segmented region.

see that the abnormal area is having a higher average

Asymmetry analysis is then performed using different

temperature of around 1°C. Any value above the

statistical parameters. The experimental results show that

threshold is marked as 1 and values less than threshold

there is a significant difference in these parameters for

makes zero. With this logic, areas of highest temperature is

healthy and sick subjects and thus they can be used for

extracted. Finally, the output matrices generated by body

detection of breast abnormality. Figure 8 shows the

boundar y identification algorithm and highest

process of segmentation.

temperature area extraction algorithm are combined

2.5 Feature Extraction for Asymmetry Analysis

together to generate a processed image. Figure 7 (a) and 7 (b) show the output of hotspot detection and final combined image. The processed image is easy to interpret and understand.

Feature Extraction is performed on segmented thermograms which includes a comprehensive set of features like Skewness, Kurtosis, Median, Variance, Standard Deviation, and Mean to completely study the

2.4 Segmentation

nature of temperature distribution in Region of Interest

The breast thermograms are segmented in Left and Right

(ROI).

breast manually using crop function in SMARTVIEW which

3. Results & Discussion The results are treated as 'Normal' case if the Hotspot does not appear in any of the breasts OR it is present in both the breasts symmetrically. If the Hotspot is present in areas like armpit, neck and inframammary folds where the temperature is naturally expected to be more, then also it is considered as 'Normal'. The thermogram is considered as 'Abnormal' if the hotspot appears only in any one of the breast part.

(a)

(b)

Figure 6. (a) Background Detection, (b) Edge Detection

(a) Figure 5. System Flowchart

24

(b)

Figure 7. (a) Hotspot Detection, (b) Final Output

No. 4 l December 2016 - February 2017 i-manager’s Journal on Pattern Recognition, Vol. 3 l

RESEARCH PAPERS ARTICLE that the thermal distribution of ROI for Normal Case is almost symmetrical in both the breasts as compared to the distribution of ROI for Abnormal Case which is distinctly asymmetrical in Left & Right breast as shown in Figure 11 (b). 3.4 Results of Feature Extraction The statistical parameter values variy between contralateral breasts in case of abnormality and is nearly same for symmetric cases. A graphical representation of the parameters is plotted for more precise presentation of the evaluated parameters and for showcasing the symmetry between the contralateral breasts. The Figure 8. Process of Segmentation

horizontal axis of the graphs represents the different statistical parameters, whereas the vertical axis represent

3.1 Abnormal Case Breast Thermograms

its values. Figure 12 (a) shows the graph of the first three

Highest temperature area appears only in left part of the

parameters, i.e. Skewness, Kurtosis, and Median. Figure 12

breast and not in right part of the breast as shown in Figure

(b) shows the graph of the remaining three parameters i.e.

9 (a) and (b). The other hotspots present at neck, armpit

Variance, Std. Deviation and Mean. Figure 12 (a) and (b)

and centre of breast are to be ignored since the

shows the graphs of parameters for normal thermograms.

temperature of these areas is naturally high. In both

Thus, almost no deviation in the values of the contralateral

thermograms, highest temperature area appears only in

breasts is clearly observed from the graph. Figure 13 (a)

any one of the breasts, which is a sign of asymmetry, and

and (b) presents the graphs for abnormal case.

thus treated as abnormal cases.

Thus, the deviation in the values of the contralateral

3.2 Normal Case Breast Thermograms

breasts is clearly observed from the graph.

Highest temperature area appears below left and right

Conclusion

parts of the breast, because of folds across the breast as shown in Figure 10 (a) and (b). No hotspot is present in the breast area. Symmetry exists in appearance of highest temperature area of both left and right parts of the breast, which implies that these are normal case thermograms.

This work presents an approach which deals with analysis of breast thermograms based on hotspot identification for abnormality detection. Experimental results of Hotspot detection show that the proposed approach is able to accurately detect the location and shape of the hot

3.3 Histogram

region in the thermogram. The presence of such hotspot

Feature Analysis is based on Histogram Generation.

is an indication of lump and needs further follow-up.

Figure 11 (a) shows the histogram of Normal Case. It shows

Asymmetry analysis based on statistical features supports

(a)

(b)

Figure 9. (a) Abnormal Case 1, (b) Abnormal Case 2

(a)

(b)

Figure 10. (a) Normal Case 1, (b) Normal Case 2

No. 4 l December 2016 - February 2017 i-manager’s Journal on Pattern Recognition, Vol. 3 l

25

RESEARCH PAPERS ARTICLE

(a)

(a)

(b) (b) Figure 11. (a) Histogram of Normal Case, (b) Histogram of Abnormal Case

Figure 10. Statistical Parameters Representation for Sick Case (a) Skewness, Kurtosis and Median Parameters (b) Varience, Std Deviation and Mean Parametrs

the abnormality detection. The results are validated by a doctor's diagnosis and are cross checked by camera software where individual pixel temperatures can be monitored. It is difficult to identify deep tumors since they don't show remarkable skin temperature difference in image. To improve decision making in all such type of cases, more data points are necessary for drawing relevant conclusions. Acknowledgment

(a)

The author would like to thank Dr. Shekhar Kulkarni for his valuable guidance and support during image acquisition. References [1]. Kapoor, P., & Patni, S. (2012). Image segmentation and asymmetry analysis of breast thermograms for tumor detection. International Journal of Computer Applications, 50(9) 40-45.

(b) Figure 12. Statistical Parameters Representation for Healthy Case (a) Skewness, Kurtosis and Median Parameters, (b) Varience, Std Deviation and Mean Parameters

26

[2]. Khandpur, R. S. (1992). Handbook of biomedical instrumentation. Tata McGraw-Hill Education.

No. 4 l December 2016 - February 2017 i-manager’s Journal on Pattern Recognition, Vol. 3 l

RESEARCH PAPERS ARTICLE [3]. Kuruganti, P. T., & Qi, H. (2002). Asymmetry analysis in

its clinical application. Biomedical Signal Processing and

breast cancer detection using thermal infrared images. In

Control, 3(4), 312-318.

th

Engineering in Medicine and Biology, 2002. 24 Annual

[7]. Umadevi, V., Raghavan, S. V., & Jaipurkar, S. (2010,

Conference and the Annual Fall Meeting of the

November). Interpreter for breast thermogram

Biomedical Engineering Society EMBS/BMES

characterization. In Biomedical Engineering and

Conference, 2002. Proceedings of the Second Joint (Vol.

Sciences (IECBES), 2010 IEEE EMBS Conference on (pp.

2, pp. 1155-1156). IEEE.

150-154). IEEE.

[4]. Mejía, T. M., Pérez, M. G., Andaluz, V. H., & Conci, A. (2015,

[8]. Wakankar, A., Suresh, G. R., & Ghugare, A. (2014,

July). Automatic Segmentation and Analysis of Thermograms

January). Automatic Diagnosis of Breast Abnormality

Using Texture Descriptors for Breast Cancer Detection. In

Using Digital IR Camera. In Electronic Systems, Signal

Computer Aided System Engineering (APCASE), 2015 Asia-

Processing and Computing Technologies (ICESC), 2014

Pacific Conference on (pp. 24-29). IEEE.

International Conference on (pp. 145-150). IEEE.

[5]. Ng, E. K. (2009). A review of thermography as

[9]. YK Ng, LN Ung, FC Ng, LSJ Sim, E. (2001). Statistical analysis

promising non-invasive detection modality for breast

of healthy and malignant breast thermography. Journal of

tumor. International Journal of Thermal Sciences, 48(5),

Medical Engineering & Technology, 25(6), 253-263.

849-859.

[10]. Zare, I., Ghafarpour, A., Zadeh, H. G., Haddadnia,

[6]. Tang, X., Ding, H., Yuan, Y. E., & Wang, Q. (2008).

J., & Isfahani, S. M. M. (2016). Evaluating the thermal

Morphological measurement of localized temperature

imaging system in detecting certain types of breast tissue

increase amplitudes in breast infrared thermograms and

masses. Biomedical Research.

ABOUT THE AUTHORS Asmita T. Wakankar is currently pursuing her Ph.D at Sathyabama University, Chennai. She has done her ME (Biomedical Instrumentation) from Govt. College of Engineering, Pune in 2002. She is working as an Assistant Professor at Instrumentation & Control Department, Cummins College of Engineering, Pune. She has published 25 papers in journals and conferences. Her areas of interest are Biomedical Instrumentation and Image Processing.

Dr. S. Lalithakumari is working as an Associate Professor in the Department of EIE of Sathyabama University, Chennai. She has completed graduation from Bharathidasan University, Thiruchirappalli (1993) in Electronics and Communication Engineering Discipline. She obtained her post Graduation in Power Electronics from Bharathidasan University, Thiruchirappalli (1999). She obtained her Ph.D from Sathyabama University, Chennai (2014). She has teaching experience of about 18 years. She has published around 30 papers in reputed journals and has filed 5 patents so far. Her research interest includes Non Destructive Testing, Image & Signal Processing and Soft Computing.

Dr. G.R. Suresh is currently working as a professor in Rajlakshmi Institute of Technology, Chennai. He received his Bachelor of Engineering Degree in Electronics and Communication from Manonmaniam Sundaranar University in 1997, Master of Engineering degree in communication systems from Madurai Kamaraj University, Madurai in 2000 and PhD degree in the Faculty of Information and Communication Engineering from Anna University, Chennai in 2010. He has more than 19 years of experience in teaching at under graduate and graduate level. He has published more than 62 Research papers in Journals and Conferences. He has received the Computer Engineering division prize from the Institution of Engineers (India) for the year 2009. He is an active reviewer in the journals, IET image Processing and International journal of Electronics. His research area includes Medical Signal & Image processing, Speech processing and WSN in Telemedicine. He is a member of IEEE, IET and life member of ISTE.

No. 4 l December 2016 - February 2017 i-manager’s Journal on Pattern Recognition, Vol. 3 l

27