Development and Validation of MRI Imaging Biomarkers for Prostate Cancer Using Deep Learning

PURPOSE/OBJECTIVE(S)

Given the increasing number of treatment options for patients with localized prostate cancer (PCa), there is a need for biomarkers to aid in risk stratification. Specifically, novel biomarkers can aid in the identification of high-risk phenotypes among similar patients in traditional NCCN risk groupings. One promising area for development is using pre-treatment prostate MRI to identify imaging biomarkers to identify prostate cancer patients at highest risk for recurrence. We hypothesized that deep learning could be leveraged to identify imaging biomarkers of aggressive PCa from pre-treatment prostate MRIs.

MATERIALS/METHODS

Our study included 1,020 patients treated at our institution between 2010-2022. Given pathologic extraprostatic extension (EPE) and seminal vesicle invasion (SVI) are associated with higher risk of treatment failure, we hypothesized that deep learning models which identified radiographic EPE and SVI would provide non-invasive imaging biomarkers associated with PCa prognosis. We trained two separate deep learning models using convolutional neural networks to predict SVI and EPE respectively. The model inputs were T2W prostate MRIs (n = 894) and models consisted of 8 convolutional layers. Dropout, L2 regularization, and data augmentation were used to improve model generalizability and reduce overfitting. Discriminatory ability of each model was measured using AUC on a blinded external test set of 221 patients. To assess the clinical utility of our imaging biomarkers, log-rank tests were used to evaluate biochemical free survival (BFS) for patients classified as high risk to patients classified as low risk. Biochemical failure was defined as post-treatment PSA >0.1 for patients who underwent radical prostatectomy (RP) or PSA >2ng/ml above nadir for patients receiving radiation therapy.

RESULTS

Within our cohort of 1,020 patients the median age was 66 with a median follow up of 4 years. 49.3% (n = 503) underwent RP and 50.7% (n = 517) received EBRT. 4% (n = 41) were low risk, 62.4% (n = 636) were intermediate risk, and 33% (n = 337) were high risk based on NCCN criteria. Deep learning models showed good discriminatory ability for both EPE (AUC 0.66) and SVI (AUC 0.74). Both imaging biomarkers showed prognostic ability to identify high risk prostate phenotypes. Patients deemed high risk based on EPE classifier had worse BFS (median 5 vs 9 years, p<.001). Similarly, patients classified as high risk based on SVI also showed worse BFS (median 5 vs 9 years, p = 0.024). Among intermediate risk patients, EPE biomarker showed an ability to identify high risk phenotypes (median 6 vs 9 years, p = 0.024).

CONCLUSION

Deep learning classifiers of prostate MRIs demonstrated the ability to stratify high-risk prostate cancer phenotypes beyond traditional risk paradigms. Imaging biomarkers represent a non-invasive method to help aid in the personalization of treatment for patients with localized prostate cancer and identify patients who potentially require treatment escalation.

Profile of osteoarticular tuberculosis in children

OBJECTIVE

To determine clinical profile of osteoarticular tuberculosis (TB) in children.

METHODS

Cross-sectional analysis from 2007 to 2013. All patients diagnosed with bone TB, spinal TB or TB abscesses were included.

RESULTS

Out of 1318 children with TB, 39 (2.96%) had osteoarticular TB, of which 16 (42%) had osteomyelitis, 8 (20.5%) had spinal involvement, 7 (17.9%) had TB synovitis, 2 (5.1%) had psoas abscess and 6 (15.4%) had abscesses. The mean age of presentation was 7.1 ± 3.5 years (range 2-14 years). Of the 33 cases in which a culture was done, 25 (64%) showed a positive culture. Drug sensitivity tests were done in 21 patients of which 10 (47.6%) tested were drug resistant, of which 4 (36.4%) were multidrug resistant (MDR), 2 (18.2%) were extensively drug resistant (XDR), 3 were pre-XDR (27.3%) and 1 was polyresistant (9.1%). Nine (23.1%) patients had TB in the past with a treatment duration of 8.3 ± 5.3 months. Contact with a TB patient had occurred in 10 (25.6%) cases. Associated pulmonary TB were seen in 6 (15.39%) and TB meningitis were seen in 1 (2.6%) patients. Surgical intervention was needed in 11 (28.2%) patients of which 5 (45.5%) underwent curettage, drainage was done in 1 (9.1%), arthrotomy in 4 (36.4%) and spinal surgery in 1 (9.1%) patient.

CONCLUSION

Drug resistant osteoarticular TB is an emerging problem in children.

Measuring change in arm hand skilled performance in persons with a cervical spinal cord injury: responsiveness of the Van Lieshout Test

OBJECTIVES

To asses whether the Van Lieshout Test (VLT) is responsive to measure changes in arm hand skilled performance (AHSP) over time during active rehabilitation in persons with cervical spinal cord injury (C-SCI) and in different subgroups of persons with C-SCI according to lesion completeness and to lesion level.

STUDY DESIGN

Longitudinal cohort study.

SETTING

Spinal Cord Injury Units in eight rehabilitation centres across the Netherlands.

METHODS

In the present study, data from a national SCI cohort study are used. Data from the VLT, the Grasp Release Test (GRT), the Functional Independence Measure (FIM) and the Quadriplegia Index of Function (QIF) were recorded at three moments during active rehabilitation: at the start of active rehabilitation (t1), 3 months after t1 (t2) and at discharge (t3). Data have been analysed using three responsiveness measures, that is, the standardized response mean, the effect size and comparison of test scores measured at different times.

RESULTS

The VLT is responsive for the intervals t1-3, t1-2 and t2-3. The VLT can be used to measure changes in AHSP in persons with a C-SCI with an American Spinal Injury Association Impairment Scale (AIS) A or B score or an AIS C or D score, as well as in persons with a C3-C6 lesion or a C7-T1 lesion. The responsiveness of the VLT is significantly correlated to the GRT, but not to the FIM and the QIF.

CONCLUSION

The VLT is responsive in measuring changes in AHSP during rehabilitation in persons with C-SCI.

Rectus femoris surface myoelectric signal cross-talk during static contractions

The clinical application of EMG requires that the recorded signal is representative of the muscle of interest and is not contaminated with signals from adjacent muscles. Some authors report that surface EMG is not suitable for obtaining information on a single muscle but rather reflects muscle group function [J. Perry, C.S. Easterday, D.J. Antonelli, Surface versus intramuscular electrodes for electromyography of superficial and deep muscles. Physical Therapy 61 (1981) 7-15]. Other authors report however, that surface EMG is adequate to determine individual muscle function, once guidelines pertaining to data acquisition are followed [D.A. Winter, A.J. Fuglevand, S.E. Archer. Cross-talk in surface electromyography: theoretical and practical estimates. Journal of Electromyography and Kinesiology 4 (1994) 15-26]. The aim of this study was to determine whether surface EMG was suitable for monitoring rectus femoris (RF) activity during static contractions. Five healthy subjects, having given written informed consent, participated in this trial. Surface and fine wire EMG from the rectus femoris and the vastus lateralis (VL) muscles were recorded simultaneously during a protocol of static contractions consisting of knee extensions and hip flexions. Ratios were used to quantify the relationship between the surface EMG amplitude value and the fine wire EMG amplitude value for the same contraction. The results showed that hip flexion contractions elicited RF activation only and that knee extension contractions elicited fine wire activity in VL only. When the relationship between RF surface and RF fine wire electrodes was compared for hip flexion and knee extension contractions, it was observed that for all subjects, there was a tendency for increased RF surface activity in the absence of RF fine wire activity during knee extensions. It was concluded that the activity recorded by the RF surface electrode arrangement during knee extension consisted of EMG from the vastii, i.e., cross-talk and that vastus intermedius was the most likely origin of the erroneous signal. Therefore it is concluded that for accurate EMG information from RF, fine wire electrodes are necessary during a range of static contractions.

Is rectus femoris really a part of quadriceps? Assessment of rectus femoris function during gait in able-bodied adults

There is conflicting evidence as to the precise activation pattern of the rectus femoris (RF) muscle during gait in able-bodied subjects. The aim of this study was to determine precise activation pattern of RF during gait in a healthy adult population. Surface and fine wire EMG from RF and vastus lateralis (VL) were recorded simultaneously at different walking speeds in five able-bodied subjects. The results of fine wire EMG showed that, at normal walking speed, RF was active only during the stance-to-swing transition. At faster speeds, increased levels of RF activity were seen during the same transition period. At the fastest speed, there was some activity during terminal stance, in addition to that at stance-to-swing transition, that could be attributed to rapid hip extension and thereby a rapid stretch of RF. However, the RF surface EMG signal was found to exhibit a bi-phasic pattern whilst walking at all different self-selected speeds. A burst of EMG activity recorded at initial contact on the surface signal but not on fine wire EMG, was due to cross-talk from vastus intermedius (VI). A close similarity was observed between the surface EMG patterns of VL and RF. VL surface and fine wire EMG profiles were identical for all the self-selected speeds with a main peak of activity beginning in terminal swing and ceasing in mid-stance. We conclude that RF is active only during stance-to-swing transition and the activity during swing-to-stance transition, as described in the literature, is very probably due to cross-talk.

A feasibility study of remote consultation to determine suitability for surgery in stroke rehabilitation

We studied knowledge transfer for the determination of the suitability of stroke patients for a specialist surgical procedure (split anterior tibial tendon transfer). Gait analysis data from patients at a general hospital were discussed with an expert in another country using personal computers, an ISDN connection (128 kbit/s) and TCP/IP-based communication tools. The key issue was whether the staff in the general hospital became better able to determine suitability for surgery. Twelve patients were studied. In three of the first four cases the advice of the remote expert changed the plan for surgery. After that the treatment plans did not change after consultation. After eight cases the local clinicians did not need to ask for further advice. There was a rapid increase in skill in determining suitability for surgery. The experience and skills of the local clinicians were thought to increase more rapidly than would have been the case without the consultations with a remote expert.

Application of a dual channel peroneal nerve stimulator in a patient with a "central" drop foot

Dropped foot is a common mobility problem amongst patients after a cerebro vascular accident. The condition arises from paresis of the muscles that control the foot movement during the swing phase of gait. If the abnormal movement is not compensated for, it results in a significant decrease in the mobility and hence quality of life. Compensation for the drop foot can be achieved through the application of functional electrical stimulation. To date, in the clinical environment, the stimulation has been applied through electrodes placed on the skin over the common peroneal nerve, and using a single channel implant device. It is well known that with these techniques it is difficult to establish a balanced response of the foot. An implantable dual channel system for stimulation of the deep and superficial peroneal nerve has now been developed for patients with a drop foot following a stroke. By stimulation of the two branches of the common peroneal nerve separately it is possible to achieve a precisely balanced dorsal flexion and eversion of the foot. Stimulation occurs via small bipolar electrodes which are placed subepineural. After successful tests on animals we have now started the two channel peroneal nerve stimulator implantation in patients. The preliminary results of the first implants are presented.

Lumbar and lumbosacral tuberculous spondylodiscitis in adults

We have reviewed, retrospectively, 66 adult patients who were treated for lumbar or lumbosacral tuberculosis. A total of 45 had a paravertebral or epidural abscess, 24 had clinical instability and 18 presented with a radiculopathy, of which six also had a motor deficit. The diagnosis was usually made on clinical and radiological grounds and they were followed up until there were clinical and radiological signs of full recovery.

Conservative treatment with antituberculous drugs was successful in 55 patients (83%). None had persistent instability, radiculopathy or neurological compromise. We feel that tuberculous spondylodiscitis, especially in the lumbar spine, can usually be satisfactorily managed conservatively and that there are few indications for surgical treatment.

Lumbar and lumbosacral tuberculous spondylodiscitis in adults. Redefining the indications for surgery

We have reviewed, retrospectively, 66 adult patients who were treated for lumbar or lumbosacral tuberculosis. A total of 45 had a paravertebral or epidural abscess, 24 had clinical instability and 18 presented with a radiculopathy, of which six also had a motor deficit. The diagnosis was usually made on clinical and radiological grounds and they were followed up until there were clinical and radiological signs of full recovery. Conservative treatment with antituberculous drugs was successful in 55 patients (83%). None had persistent instability, radiculopathy or neurological compromise. We feel that tuberculous spondylodiscitis, especially in the lumbar spine, can usually be satisfactorily managed conservatively and that there are few indications for surgical treatment.

Sacral anterior root stimulation for bladder control: clinical results

The Brindley bladder stimulator delivers intermittent stimulation to the anterior sacral roots. The stimulus parameters can be adjusted and set specifically for individuals. Its primary purpose is to improve bladder emptying, thereby to eliminate urinary infection and to preserve kidney function. It also assists in defecation and enables male patients to have a sustained full erection. In our unit so far 38 patients with a complete spinal cord lesion have received a Brindley bladder stimulator implant. One patient died 2 weeks after the surgery due to pulmonary embolism. Two other patients died due to unrelated causes during the follow up period. They used their implants for less than 1 year. Results relating to these 2 patients and the remaining 35 patients who regularly use their implant are presented. The follow-up period ranged from 3 months to 12 years. Residual urine volumes are substantially reduced in all patients; in 24 patients the residual urine volume is less than 30 ml. All patients have increased bladder capacity. Thirty-one patients are continent. Out of 33 males 29 can achieve a sustained full erection using the stimulator. Twenty-seven patients use the implant for bowel function. The following complications were encountered: (1.) Cerebro-spinal fluid collection occurred around the implant in 3 patients during the post-operative period; (2.) Receiver failure occurred in 3 patients. A successful replacement with a new receiver block was carried out in these cases. It is concluded that the use of a bladder stimulator in selected patients gives long term favourable results.

Assessment of rectus femoris function during initial swing phase

The normal human gait cycle is divided into two phases, namely, stance and swing. The objective of stance is to provide support, stability and propulsion and that of swing is to provide ground clearance and limb advancement. Knee flexion is essential during swing to lift the foot off the ground for limb advancement. The complex mechanisms involved in producing limb advancement can produce excessive knee flexion at faster walking speeds. Under these circumstances the shank needs to be decelerated to reduce the amount of knee flexion. It is assumed that rectus femoris (RF) is active for a very short period at the beginning of the swing phase (Perry J. Gait Analysis-Normal and Pathological Gait. Slack Incorporated, USA, 1992; Scott L, Ringwelsky D, Carroll N. Transfer of rectus femoris: effects of transfer site on moment arms about the knee and hip. J Biomech 27;1994:1201-1211) and the amount of this activation is proportional to the walking speed and thus to the generated knee moment and the angular acceleration of the lower limb segments. However, there is very little evidence to support these assumptions. The objective of this study was to study this relationship. Quantified electromyogram of RF and vastus lateralis (VL), using surface electrodes, were examined, body mounted kinematic sensors such as seismic accelerometers and gyroscopes were used to measure segments' angular accelerations and the net muscular knee torque calculated from the kinematics of the segments at various speeds. The results showed that RF and VL work independent of each other during the initial swing phase. The amount of RF activity is clearly related to walking speed. The muscle activity increases with increasing walking speed. The relationship between the angular acceleration of the shank and the amount of RF activity is linear. The active knee moment, as a function of the shank's angular acceleration, shows the same high correlation to the EMG signal of RF.