Cancer MRI Medical Image Analysis and Detection System (MIAS) 60.17 Screenshots

 

 

Status Report

Study Number: 07-010

Study Title: Establishing Accuracy of Medical Imaging Analysis Software (MIAS) in Identifying Cancerous Growth in MRI, CT Scans and PET Scans
Approval Date: June 15, 2007

IRB Expiration Date: June 14, 2008

 

 

1. What is done:

 

            The medical study evaluating the MIAS software was approved on June 15^th, 2007. From June 15, 2007 to September 2007, research planning sessions were held to develop a methodology for research. Based upon the research discussion, it was decided that we would evaluate the MIAS software using cranial MRI scans using the ?3 Plance Loc? procedure. From October 2007 to January 2008, 25 of 300 images were used to calibrate the MIAS software. From February 2008 to present, April 2008, the first and second part of the double blind analysis of the MRI images was performed with 150 of 275 to be completed by April 28^th.

            The initial results of the calibration study from December 2007 are:

 

Two protocols for making a determination on ?3 Plane Loc? series were evaluated. In addition, the evaluation considered the accuracy of a single image determination versus a series determination. The probability of significance is set at p<=0.15, which is the naked eye misreading rate as mentioned in past meetings. The first test was a one-way anova comparing the statistical difference between a one reading, determination and a series determination. The one reading determination, for n=169, has an accuracy rate of 97% (164/169) with a false positive rate of 3% (5/169) , and a false negative rate of 0%. The series has an accuracy rate of 70% (21/30) with a false positive rate of 30% (9/30, two studies), and false negative rate of 0%. When performing the one-way Anova, there is a statistical difference between the single image determination and a series determination at p=0.11, which is significant to p=0.15. The 70% success rate is mathematically determinable by root of 0.70 to 15th power which equals 0.976 or 97.6% success rate. The second test, comparing two protocols 2 and 3, there was not a significance difference according to the null hypothesis of p=0.15. The value of p=0.58 is not significant to the null hypothesis. But, between images, there was a significance of p=0.07, which means that the protocols are sensitive to the images. In test 3, a two-way Anova was done to compare if there was difference between the sample or the determination method used, as the series consensus determination of protocol 3 versus the consensus protocol 2.  The analysis showed the results for both image series difference and consensus determination were not significant as their p values were 0.50 and 0.36 respectively. In the two protocols used, they are similar to each other in accuracy. In test4, a two-way Anova was done with replications to determine if there is significance between the series id or the consensus determination versus the single MRI determination. There is a significance between the determination method where the p=3.8e-7, but there is not a significance between the series id where p=0.50. What this means is that single image determination has a greater accuracy than the consensus determination. In the single image determination, one result of the MIAS is that it is able to separate the three positive areas, and determine correctly which is the false positive areas (2) and the true positive area (1).

 

            The current results from the ongoing double blind analysis, n(series)=25 and n(images) = 423 are:

Wheaton Medical Study 07-010                                                                                                                       24-Mar-08                                                                                                                   Using the MRI ?3 Plane Loc? protocol.                                                                         Summary          Series Correct  1          25        Incorrect          0          Total    25            Percentage       100.00%          False Pos         0.00% Images Correct 1         414      Incorrect          9          Total    423      Percentage            97.87%            False Pos         2.13%

 

The series accuracy is increased to 100%, n=25, and a false positive rate of 0.00%. For individual images, the image diagnosis accuracy is 97.87%, n=423, and a false positive rate of 2.13%. Since the initial, calibration study of December, the series diagnostic accuracy has increased from 70% to 100%, and the individual image, diagnostic accuracy has increased from 97.6% to 97.9% with the false positive rate reduced from 2.40% to 2.14%.

            In addition, since the calibration study of December 2007, we have made the program easier to use, and using the current DICOMM imaging program supplied with the images, we can process 5-7 patient series per hour. When other DICOMM imaging programs are used, the processing rate could be increased to 15-21 patients per hour based upon using a single-core computer with a 1.8 GHz processor.

 

 

2. What is being planned:

            We are expecting to complete the IRB study by August 14, 2008 (Amendment for extension is included as attachment).  The reason we are asking for this extension is so that we can prepare a report and research paper.

 

 

3. What problems we have encountered:

 

 

            The main problems we encountered were in the early phases. The program was modified to make it easier for the future oncologist or oncologist technician to use. Once the user interface was changed, it became a three step process to analyze a patient series.

 

4. Conclusion:

 

            The MIAS application is exceeding our initial expectations, and we have high confidence that it can be used as primary diagnostic tool. Also, we have learned how to improve the program and what gained new basic research insight into the development and detection of cancer.

 

 

 

 

 

 

 

 

 

 

David Kanecki, MBA, A.C.S., Bio. Sci. 

 

 

FIRST RESULTS OF MEDICAL STUDY USING MIAS 60 LABORATORY 

David Kanecki, MBA, A.C.S., Bio. Sci. 

Kanecki Associates, Inc., * P.O. Box 866 * Kenosha, WI 53141 * UNITED STATES 

david@kanecki.com 

ABSTRACT Based upon initial testing, the MIAS 60 program could be used as a screening program for cancer, and as an assistant for cancer determination. As a screening program, the program has a 95.6% accuracy rate, and as a cancer determination program, it has a 82.4% accuracy rate. In addition, there are no false positives given as a result. This means that a patient will not be misdiagnosed because of false positive. The screen accuracy range is approximately 92-100%, and the cancer detection range is 65-100 percent. In screening, the program surpasses a human reader, and in cancer detection via a ?3 Plane Loc? analysis, it is approximately equal to a human reader. *

 

 This study was performed at Wheaton Franciscan Hospital of Racine, WI. 

INTRODUCTION The purpose of this study is to determine the accuracy rate of the MIAS 60 program to detect cancer in patients using a ?3 Plane loc? analysis. The study consists of analyzing 300 patient samples, and this report gives results for the first set of 25 patient samples. The MIAS 60 Laboratory program provides an analysis using multiple measurements and generating a single conclusion for an individual slide and the whole series as positive or negative. The conclusion from the whole series is what is important in determining if someone has cancer. The first part of the testing involved calibrating the MIAS 60 Laboratory program along with establishing patient control sets to use a reference. These control sets were selected by the variation of the ?3 Plane Loc? process and on comparative anatomy of the scans to provide a close, accurate starting point. Once the control sets were setup, the first experiments were done.

 

** MATERIALS AND METHODS The first part of performing the medical analysis was to select a set of patients as the reference set for the study. The selected set is references by a notation of gender (f/m), nc (negative control) or pc (positive control), and nc or pc number, i.e. 1. Thus, ?fnc1` would indicate that the a female, negative, control one was used for the analysis. The exam use specifies which reference set to use, based upon the variation of the ?3 place loc? examination performed. When there is a difference between the predicted result of cancer and the expected result of cancer, a second review is conducted. The second review consists of comparing the patient report for the test, and determining if the symptoms are positive or negative, relative to the two patient samples. A positive designation is given when a patient or patients have symptoms found in two different anatomical areas of the brain. This method is used because all of the patient samples we are based upon the patient having symptoms as left sided weakness, dizziness, blurred vision, etc. Thus, the second review is used to compare conflicting results between the predicted determination by MIAS. Based upon the notation used, the following table shows which patient data we used as control sets: Patient ID Gender Exam Use Cancer Present Notation Name Number of MRI scans P010 F 1 P Fpc1 21 P001 F 1 N Fnc1 15 P004 M 1 N Mnc1 15 P005 F 3 N Fnc3 21 (2 visits) P006 F 1 N Fnc2 42 P007 F 4 N Fnc4 33 P022 M 5 N Mnc2 15 From this sample set, a series of 25 experiments are performed. The goals of the experiments are 1) to determine the accuracy rate of predicting cancer in a MRI test series and 2) to determine the accuracy of recognizing significant, symptom areas that are not cancerous.

 

The summary of the results are: By Cancer By Symptom Total 23, 100% 23, 100% Correct 19, 82.6% 22, 95.6% False Positive 4, 17.4% 1, 4.3% False Negative 0, 0.0% 0. 0% With the cancer analysis, the program is approximately equal to a human reader, p=0.15 error rate, and the program error rate is p=0.164. *In addition, the program has a false positive rate 17.4% percent which means the reading accuracy can range from 69- 100% accuracy, exceeding a human reader at the upper levels. By having the program issue a false positive and not issue a false negative, this means the program`s error will not result in a missed diagnosis. With the symptoms, the program is better than a human reader with an accuracy rate of 95.6% and an error rate of 4.4%, p=0.044, which is better than p=0.15, our null hypothesis for a human reader. In addition the program has a 4.4% false positive rate which means that it range of accuracy is 91.2% to 100%, which is better than the null hypothesis of a human reader. Therefore, this shows the MIAS can be used for screening along with a secondary review

 

*** CONCLUSION The results of this study show that the MIAS surpass a human reader for screening, and is approximately equal to a human reader for cancer determination. Since the program does not issue a false positive, the benefit is toward better treatment because a lack of a false negative diagnosis means that a patient is not misdiagnosed where a cancer exists. Based upon these results, the MIAS program could be used as a screening program for cancer, and as an assistant for diagnosing cancer. The patient samples used in this study were supplied by Wheaton Franciscan Hospital of Racine, WI. The medical study is being conducted at the same hospital.

 

MIAS, Medical Image Analysis System, uses a two dimensional approach to spot cancer. The reason this approach is more accurate and efficient than the three-dimensional approach used today is due to the computers software ability for spotting the cancers location by pixel rating as opposed to someone having to determine an area from a visual contrast. After the individual determines which areas that the fluorescent dyes highlight an assumption is made which runs a risk of human error. The ability to spot the cancer by computer software pixel ratings increases location accuracy and reduced errors in reading radiological CT, MRI, and PET scans. Combined the two test run consistently could solve the problems that exist in radiological diagnosis!

 

** The Medical Imaging Analysis Software uses a special patented Dicomm File Converter to extract the bipmap files used by all Medical Scanners. Then, a pixel is assigned a value of 0 (black) ? 24 MILLION (bright white). A cancer usually is observed on a scan with a value of 150 or over Based upon the statistical value and pixel value a cancer can be detected to an exact location on our grid of 5000 by 5000 pixels The MIAS provides an image of the cancer shape, location, intensity, and many other features to detect cancer smaller than 1mm in size!

 

*** Imaging Enhancements have the ability for earlier cancer detection less than 1mm sq from modifying the pixel rating from 0-255 to 0-24 million range. This has been done when the program applies ability while reading jpeg files or bitmap files with 24-bit color. Currently, the program works on the 2006 Dicomm standard of a bitmap file of 8-bit grayscale and 24-bit gray scale files that is run on all machines.

 

 

 

 

 

 

 

 

A Comparative Analysis of the Medical Image Analysis System 60 (MIAS)

 

To Various Computer Aided Detection (CAD) Applications

 

 

November 5, 2008

 

David Kanecki

MBA, A.C.S., Bio. Sci.

Doctor of Management

Doctoral Candidate

Status of Computer Aided Detection, Version 1.0

Current systems have as weakness that they cannot they cannot make a diagnosis. The current system can only provide viewing capability. The benefit of the ?MIAS? program is that the computer can provide diagnosis. The result of this is that it can be used more hospital areas, including smaller community hospitals. Thus, the ?Second Opinion? allows for an evaluation and diagnosis by computer software which is based on using specific criteria similar to what an a physician may use..

Current  systems are CAD 1.0 systems meaning that they still rely on the physician`s evaluation completely. In essence, they are computer viewer`s of MRI slides similar to the light viewer used to view X-Ray slides.  Our MIAS is a complete system where the computer actually reads, evaluates, charts, and makes a diagnosis of the MRI for a patient being evaluated for cancer. In addition, it provides quantitative information so that the it can be charted as compared to evaluating a MRI by using the ?naked? eye and guessing changes in cancer. Our program is that only program that makes a firm diagnosis and provides a reliability index of that diagnosis. The disadvantage of CAD 1.0 systems is  that they are still viewers. Although, they use computer graphics, they are still viewers not analyzers and cannot give a diagnosis. To perform better the MIAS is needed to use the ability of the computer and software to help make a decision and make a diagnosis which is a CAD 2.0 system. Our MIAS system is a CAD 2.0 system. The benefit of a CAD 2.0 system is that it the computer makes a diagnosis so it aids the physician in verifying what they have determined, and in certain cases where an expert physician is not available, it can help to make a screening determination for further examination. Thus, the CAD 2.0 system is an active tool that makes a diagnosis as compared to showing an image and leaving the evaluation to the physician viewing it. With CAD 2.0, the physician`s expertise and time are utilized better because it can the physician in providing detailed analysis along with a specific diagnosis. 

 

In order to facilitate many cost-effective cancer screening stations, it will be advantageous to do the screening on a remote computer. The reason why a screening on a remote computer is needed is because it can be setup in a non dedicated MRI area. The benefit of this is that a special room is not needed as for the MRI setup used currently. For example, when an MRI is performed on a patient, the MRI scans the patient. Then, the physician must evaluate the scan on the computer that is directly used by the MRI instrument. The result of this is that you can only scan a patient or evaluate a patient scan at one time. Thus, the MRI cost increase because you using a specialized area to evaluate a scan. With screening on a remote computer, the cost are lower, because a standard hospital area can be used as opposed to a specialized MRI area. In addition, the remote screening can be done in more areas of the hospital than compared the one fixed area of the hospital used for MRI evaluation. The MIAS program is designed to work with or without a database. The database feature is added to allow results to be charted quickly and reduce the amount of paper needed as the results are directly entered into a hospital patient records system. The MIAS can run on a PC for the evaluation of the MRI scan, but the MIAS does not have the capability to access the control mechanism of various MRI instruments. The MIAS is an evaluation tool. By using the MIAS, the gaps filled are the ability to determine if treatment is needed sooner,  ability to have a second computer diagnosis which increases the physician`s productivity and quality because the MIAS will specifically show and provide a report where the cancer areas are located in a easy format,  ability to provide greater treatment with the current facilities available because the MIAS can evaluate the scan in a standard hospital area as opposed to a specialized area like the MRI area which means that all parts of the hospital can operate at a greater efficiency and quality to the patient and on timely basis

 

Evaluation of Claims of Using MIAS

In reviewing various studies of using computer aided detection for mammography and MR, I have found that the MIAS program in MR use and in mammography use is better individual and series diagnosis in terms of sensitivities and false positions per image. The thesis is supported by comparing various research studies. (Roman  et al, Obenauer et al,  Shiraishi et al, van Engeland et al, and Gan et al) The sensitivity levels for a series detection ranges from 50.3% to 73% compared with our initial studies for the MIAS of 82% to 100%, and with an individual series sensitivity of others of 71% to 91% compared to 93% to 100% for our MIAS. The significant issue found is that positive detection has a high linkage to false positive areas in 53% of the cases (van Engeland et al) compared to 20% of the case with our MIAS. The number of false positive per image in various studies (Roman  et al, Obenauer et al,  Shiraishi et al, van Engeland et al, and Gan et al) for MRI range from 0.5 per image (Obenauer et al) to 1.02 with mammography (Obenaurer et al). On the MRI, 3 Plane LOC our false positive per image is approximately 0.26 per image, significant compared to 0.5 per image. With the MIAS Mammography, the initial evaluation will be completed soon.  In all of the studies examined (Roman  et al, Obenauer et al,  Shiraishi et al, van Engeland et al, and Gan et al), the individual slide sensitivity rate for MR and mammogram is higher than the sensitivity rate for a series, which is consistent with our findings from September 2007 to May 2008.

            Some of the difficulties encountered in computer aided detection are the diagnostic determination of white matter spaces (Lasbury et al, Groeschel et al). The result of white matter decreases the sensitivity rate for an individual slide to 80% (Groeschel et al). In our MIAS, we have been able to detect white and our MIAS can detect and accurately diagnose at a higher rate.94.1% and 73.3% for individual and series respectively. In individual series, our MIAS is less susceptible to white matter issues in individual scans, but it relative close to accuracy of other systems in series scans. (Groeschel et al)

            Therefore, the MIAS for MR and mammography is better in most areas. Please see the notes in the annotated references for additional discussion.

 

Detailed Discussion of Evaluation of Current Computer Aided Detection

            Lasbury et al (Lassbury, 2001) report that MRI with computer aided detection (CAD) can recognize white matter deficiencies in patients. With our MIAS program for CAD we have observed that the white matter is related to ischemia and the medical imaging analysis system (MIAS) was able to correct diagnose the effects of the finding. Another finding by Lasbury et al is that 26% of the white matter in a MRI is related to significant findings. With the MIAS, the correlation to significance was observed in the MR experiments using the MIAS. The MIAS includes processes that address issues of white matter in CAD, and can correlate them.

            Roman et al (Roman et al, 2005) reported the results correlating CAD to accuracy in detection. In one analysis, a correlation percentage of 84% is found using the Bland-Alman analysis for MRI images. With the MIAS, The 84% series sensitivity is a figure that we used as null hypothesis due to personal communication in Sept 2007. Our expectation is the MIAS will have a greater accuracy of  84%. Another observation is Diagnostic difficulties occur to de-phasing, and with the MIAS, we have solved the issue of de-phasing by applying a third layer of area analysis. The implications are that the MIAS is advanced compared to what is shown in the literature in mass recognition issues.

            Obernauer et al (Obenaurer, 2006) review the false positive rates of various CAD programs to MRI image evaluation. One of the findings is that the False positive marker rate is 205/412 or 49.76% and further improvements are necessary for computer aid detection, or, a 50.24% accuracy rate compared to a 97% accuracy rate with our MIAS. The 97% single image accuracy rate is due to initial testing performed in the period of Sept 2007 to May 2008. In a comparison of false positives per image, Obenauer et al, found 0.5 false positives per image with computer aided detection with MRI or 7.0 false positives per ?3 Plane LOC? series compared with 2.0 false positives per image with our MIAS. In comparing current CAD results to a series MR evaluation, Obenauer et al, found that  77/226 or  34.1% missed or 65.9% accurate compared with our Mias is 92% accuracy for a series.

            When comparing CAD system to detecting specific masses Groeschel et al (Groeschel, 2006) , found that VRS (Virchow-Robin Spaces) can be viewed in 80% of MRI cases. The MIAS was able to detect cases of ischemia or other hard to determine findings in specific cases and to classify correctly as positive or negative. The finding detecting specific masses is from the study performed Sept 2007 to May 2008. The result is that the MIAS is consistent with other applications in identifying specific mass areas.

            With a specific study by Shiraishi et al  (Shiraishi, 2007) evaluating CAD in evaluation accuracy of MRI scans for specific areas of the body, the results of MIR bone scan analysis shows 95.3%, with 5.97 false positives per view, and 83.2% with 6.02, respectively. In our own comparative study, using the ?3 Plane LOC? analysis with our MIAS, we had 97% accuracy with a  0.26 false positives per view for a single slide and a 92% accuracy with a  0.33 false positive views per slide. Overall, Shiraishi et al, found series sensitivity rates are 62% to 100% for series and 78% to 100% for individual slides. The range shows a wide variation in where the CAD program is applied in evaluation. With our MIAS, Our lowest series rate is 82% to 100% and individual rate of 93% to 100%. This is verified by excel data collected from the study of Sept 2007 to May 2008.

            Evaluating CAD for the detection of persons with breast cancer, van Engeland et al (van Engeland, 2007), found lesion sensitivity detection rate of 61% for mammogram and 0.1 false positive images per scan. In calibration studies of the MIAS for mammography, the sensitivity rate for mammogram is approximately 80%. The result is that our MIAS is CAD for persons with breast cancer is better than what is currently reviewed in the literature.  Another observation as part of the study by van Engeland et al, is true positive sensitivity at 64% with 34% of that set linked to a false positive. The implication is that over half of the positives found are false positives. In calibration studies of the MIAS for mammogram analysis, we found the number of true positives linked to a false positive was less than 20% of the set compared to 53% of this paper`s study set. An additional result from the study by van Engeland et al, shows that with with advanced image processing, sensitivity increased to 77% to 91% range for individual slides. With our MIAS MRI, the sensitivity for individual slides is 93% to 100%.

            The study by Gan et al evaluated the use of CAD with different types of computational resources. What is found is that Small benchmark set of samples, sensitivity at 85%. This means that the smaller the sample size, the more accurate the CAD. With out MIAS regarding small benchmark size, we found studies our sensitivity is 92%, slightly higher. In large benchmark results the sensitivity obtained is 73%, when rendering a correct diagnosis, as when a patient is evaluated for a specific MR series. On large series sensitivity using ?3 Plane LOC? has a sensitive of 90% for a series which is significantly higher.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

            The advantage is that the radiologist and oncologist can see where the cancer areas and where they are spreading. In addition, a Cartesian mapping system is provided to provide a reference between scans. In the top of the image is a legend that shows the positive areas so that they can be charted for further treatment.

 

 References

Gan, X.; Kapsokalivas, L.; Skaliotis, A.; Steinhöfel, K.; Tanga